基于变长SVTR的文字识别模型及ONNX部署

使用SVTR完成文字识别模型及onnx导出

AI Studio

4147人浏览 · 2022-08-27 16:20:47

AI Studio · 2022-08-27 16:20:47 发布

基于变长SVTR的文字识别模型及ONNX部署

SVTR是2022年最新提出的一个一阶段文字识别模型，同时经过改进的版本被应用于PaddleOCRV3中。

当笔者试图使用其原始版本训练文字识别模型并进行部署时，遇到了一系列的问题。

其中最重要的是对变长文本的不支持及超长文本的识别错误问题。在GitHub的ISSUE区，该问题也被反复提及。

因此笔者在经过自己的学习后，初步解决了SVTR模型从训练推理到部署的全流程，

为了方便大家的学习及使用完成了该项目。

⭐ ⭐ ⭐ 欢迎点个小小的Fork支持！⭐ ⭐ ⭐

1.0 基于SVTR的文本检测

  title: SVTR: Scene Text Recognition with a Single Visual Model
  venue：CVPR2022
  paper: https://arxiv.org/abs/2205.00159
  code: https://github.com/PaddlePaddle

本项目主要内容如下：

对论文《SVTR: A Single Visual Model for Scene Text Recognition》进行介绍
介绍其参数及PPOCRv3中的使用。

1.1 场景文本识别介绍

场景文本识别可以看作是从图像到字符序列的跨模式映射。

场景文本识别旨在将自然图像中的文本转录为数字字符序列，该序列传达了对场景理解至关重要的高级语义。但由于文本变形，字体，遮挡，混乱的背景等方面的不同，识别任务具有极强的挑战性。

主观场景文本识别模型通常包含特征模型和序列模型。这种两阶段模型虽然准确，但效率较低。本文提出了一个用于场景文本识别的一阶段的视觉模型SVTR。仅通过单个视觉模型就完成了特征提取和文本转录这两个任务。验证了单视觉模型在合理的结构设计下，既能保证推理速度，又能获得更好的识别效果。

1.2 模型简介

上图为SVTR的结构图。SVTR仅使用单视觉模型，无语言模型，达到与主流视觉-语言两阶段模型相同或者更高的精度。同时，相比与主流的模型，速度更快，参数量更低。

由于其优越的性能，SVTR被PP-OCRv3所青睐作为其识别基础模型发布，并且得到了开发者们的广泛关注。

PP-OCRv3是百度开源的超轻量级场景文本检测识别模型库，其中超轻量的场景中文识别模型SVTR_LCNet使用SVTR作为基础结构，进行精度优化和轻量化。

详见PP-OCRv3介绍。

为了保证速度，SVTR_LCNet将SVTR模型的Local Blocks替换为LCNet，使用两层Global Blocks。

在中文场景中，经过多次优化，SVTR_LCNet的最终精度为79.4%。

具体的：

GTC：Attention指导CTC训练策略；
TextConAug：挖掘文字上下文信息的数据增广策略；
TextRotNet：自监督的预训练模型；
UDML：联合互学习策略；
UIM：无标注数据挖掘方案。

其中 UIM：无标注数据挖掘方案 使用了高精度的SVTR中文模型进行无标注文件的刷库，该模型在PP-OCRv3识别的数据集上训练，精度对比如下表。

中文识别算法	模型	精度	时间
PP-OCRv3	SVTR_LCNet(h32)	71.9.4%	6.6ms
PP-OCRv3	SVTR_LCNet(h48)	73.98%	7.6ms
+GTC	SVTR_LCNet(h48)	75.8%	7.6ms
+TextConAug	SVTR_LCNet(h48)	76.3%	7.6ms
+TextRotNet	SVTR_LCNet(h48)	76.9%	7.6ms
+UDML	SVTR_LCNet(h48)	78.4%	7.6ms
+UIM	SVTR_LCNet(h48)	79.4%	7.6ms
PP-OCRv3	SVTR_LCNet	79.4%	7.6ms
SVTR	SVTR-Tiny	82.5%	97ms

笔者按：

即使在经过多次优化后SVTR-Tiny模型的精度依然低于SVTR-Tiny，为什么选择SVTR_LCNet作为模型backbone？

这其中最明显的原因是由于SVTR-Tiny模型推理速度太慢，其实还有一个重要原因。

由于SVTR中使用了绝对位置编码，所以导致模型在推理时对长文本处理能力极差。

在PPOCR中采用的数据resize策略是对图像按照等比例进行缩放。所以推理时，长文本会直接报错。

为了预防这一报错，ppocr在模型定义时加入这一警告保证模型定义及部署为onnx时，宽度必须固定。
```
  assert H == self.img_size[0] and W == self.img_size[1], 
    f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
```
此外，在使tools/infer_rec.py时也默认导出模型的输入图像size必须为定值。

为了使SVTR能适应变长序列，笔者删除了原模型的绝对位置编码模块。并完成了模型的部署。

1.3 模型参数

在SVTR模型使用中，需要注意的是官方给出的参数为SVTR-Tiny的模型参数（在PaddleOCR/configs/rec/rec_svtrnet.yml中）

在论文中，介绍的模型有四种，使用时，将rec_svtrnet.yml配置文件中Backbone替换即可(默认输入图像的img_size为48*256，字符长度：25])：

在这里插入图片描述

笔者在这里给出这四种模型的参数及性能对比：

在这里插入图片描述

SVTR-T
Backbone:

    name: SVTRNet
    img_size: [48, 256]
    out_char_num: 25
    out_channels: 192
    patch_merging: 'Conv'
    embed_dim: [64, 128, 256]
    depth: [3, 6, 3]
    num_heads: [2, 4, 8]
    mixer: ['Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global']
    local_mixer: [[7, 11], [7, 11], [7, 11]]
    last_stage: True
    prenorm: false

SVTR-S
Backbone:

    name: SVTRNet
    img_size: [48, 256]
    out_char_num: 25
    out_channels: 192
    patch_merging: 'Conv'
    embed_dim: [96, 192, 256]
    depth: [3, 6, 6]
    num_heads: [3, 6, 8]
    mixer: ['Local','Local','Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global','Global']
    local_mixer: [[7, 11], [7, 11], [7, 11]]
    last_stage: True
    prenorm: false

SVTR-B
Backbone:

    name: SVTRNet
    img_size: [48, 256]
    out_char_num: 25
    out_channels: 256
    patch_merging: 'Conv'
    embed_dim: [128, 256, 384]
    depth: [3, 6, 9]
    num_heads: [4, 8, 12]
    mixer: ['Local','Local','Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global','Global','Global','Global','Global']
    local_mixer: [[7, 11], [7, 11], [7, 11]]
    last_stage: True
    prenorm: false

SVTR-L
Backbone:

    name: SVTRNet
    img_size: [48, 256]
    out_char_num: 25
    out_channels: 384
    patch_merging: 'Conv'
    embed_dim: [192, 256, 512]
    depth: [3, 9, 9]
    num_heads: [6, 8, 16]
    mixer: ['Local','Local','Local','Local','Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global','Global','Global','Global','Global','Global']
    local_mixer: [[7, 11], [7, 11], [7, 11]]
    last_stage: True
    prenorm: false

参考资料

二. 数据集简介

本次飞桨学习赛：中文场景文字识别竞赛数据集共包括6万张图片，其中5万张图片作为训练集，1万张作为测试集

数据集采自中国街景，并由街景图片中的文字行区域（例如店铺标牌、地标等等）截取出来而形成

所有图像都经过一些预处理，将文字区域利用仿射变化，等比映射为一张高为48像素的图片，样例如下图所示：

在这里插入图片描述

标注：魅派集成吊顶

标注文件：

数据集内容：
- 训练集图像压缩包：train_images.zip
- 平台提供的标注文件为.csv文件格式，文件中的四列分别为图片的宽、高、文件名和文字标注。样例如下：
  - 格式为：图像名称文字标注
  - 样例： 0.jpg 福
- 测试集图像：test_images.zip

三. 代码实现

第三部分对SVTR进行训练，并完成推理过程。

3.0 安装环境所需的包

!git clone https://gitee.com/paddlepaddle/PaddleOCR.git 
!pip install --upgrade pip --user
!pip install -r PaddleOCR/requirements.txt
!pip install paddle2onnx
!pip install onnx
!pip install onnxruntime-gpu

fatal: destination path 'PaddleOCR' already exists and is not an empty directory.
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Requirement already satisfied: pip in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (22.1.2)
Collecting pip
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/1f/2c/d9626f045e7b49a6225c6b09257861f24da78f4e5f23af2ddbdf852c99b8/pip-22.2.2-py3-none-any.whl (2.0 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m2.0/2.0 MB[0m [31m9.0 MB/s[0m eta [36m0:00:00[0m:00:01[0m00:01[0m
[?25hInstalling collected packages: pip
Successfully installed pip-22.2.2

[1m[[0m[34;49mnotice[0m[1;39;49m][0m[39;49m A new release of pip available: [0m[31;49m22.1.2[0m[39;49m -> [0m[32;49m22.2.2[0m
[1m[[0m[34;49mnotice[0m[1;39;49m][0m[39;49m To update, run: [0m[32;49mpython -m pip install --upgrade pip[0m
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Collecting shapely
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/a2/73/07f507b539527579c026ed5f411c35e287685ce794c4dfb7edf6675cdc24/Shapely-1.8.4-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (2.0 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m2.0/2.0 MB[0m [31m4.2 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hCollecting scikit-image
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/2d/ba/63ce953b7d593bd493e80be158f2d9f82936582380aee0998315510633aa/scikit_image-0.19.3-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (13.5 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m13.5/13.5 MB[0m [31m10.7 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hCollecting imgaug==0.4.0
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/66/b1/af3142c4a85cba6da9f4ebb5ff4e21e2616309552caca5e8acefe9840622/imgaug-0.4.0-py2.py3-none-any.whl (948 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m948.0/948.0 kB[0m [31m9.8 MB/s[0m eta [36m0:00:00[0ma [36m0:00:01[0m
[?25hCollecting pyclipper
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/5b/88/737eb2291aad95d77765b8bc6b4cb2a59b69e72b5d1922c7bab9a0979cdb/pyclipper-1.3.0.post3-cp37-cp37m-manylinux_2_5_x86_64.manylinux1_x86_64.whl (604 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m604.2/604.2 kB[0m [31m9.6 MB/s[0m eta [36m0:00:00[0ma [36m0:00:01[0m
[?25hCollecting lmdb
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/4d/cf/3230b1c9b0bec406abb85a9332ba5805bdd03a1d24025c6bbcfb8ed71539/lmdb-1.3.0-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (298 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m298.8/298.8 kB[0m [31m12.0 MB/s[0m eta [36m0:00:00[0m
[?25hRequirement already satisfied: tqdm in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from -r PaddleOCR/requirements.txt (line 6)) (4.27.0)
Requirement already satisfied: numpy in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from -r PaddleOCR/requirements.txt (line 7)) (1.19.5)
Requirement already satisfied: visualdl in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from -r PaddleOCR/requirements.txt (line 8)) (2.3.0)
Collecting rapidfuzz
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/3b/75/114ba9898e9443adaccdfde6b3593b8d03275898d41ae46d920b27aa1733/rapidfuzz-2.6.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (2.2 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m2.2/2.2 MB[0m [31m9.5 MB/s[0m eta [36m0:00:00[0m:00:01[0m00:01[0m
[?25hCollecting opencv-contrib-python==4.4.0.46
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/08/51/1e0a206dd5c70fea91084e6f43979dc13e8eb175760cc7a105083ec3eb68/opencv_contrib_python-4.4.0.46-cp37-cp37m-manylinux2014_x86_64.whl (55.7 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m55.7/55.7 MB[0m [31m6.3 MB/s[0m eta [36m0:00:00[0m00:01[0mm00:01[0m
[?25hRequirement already satisfied: cython in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from -r PaddleOCR/requirements.txt (line 11)) (0.29)
Collecting lxml
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/12/60/221cde1b87fcbc9260f92f939d99891ec151cf2ee3849ba909f02b614686/lxml-4.9.1-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_24_x86_64.whl (6.4 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m6.4/6.4 MB[0m [31m7.5 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0mm
[?25hCollecting premailer
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/b1/07/4e8d94f94c7d41ca5ddf8a9695ad87b888104e2fd41a35546c1dc9ca74ac/premailer-3.10.0-py2.py3-none-any.whl (19 kB)
Requirement already satisfied: openpyxl in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from -r PaddleOCR/requirements.txt (line 14)) (3.0.5)
Collecting attrdict
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/ef/97/28fe7e68bc7adfce67d4339756e85e9fcf3c6fd7f0c0781695352b70472c/attrdict-2.0.1-py2.py3-none-any.whl (9.9 kB)
Requirement already satisfied: matplotlib in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (2.2.3)
Requirement already satisfied: Pillow in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (8.2.0)
Requirement already satisfied: six in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (1.16.0)
Requirement already satisfied: imageio in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (2.6.1)
Requirement already satisfied: scipy in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (1.6.3)
Requirement already satisfied: opencv-python in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (4.1.1.26)
Collecting tifffile>=2019.7.26
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/d8/38/85ae5ed77598ca90558c17a2f79ddaba33173b31cf8d8f545d34d9134f0d/tifffile-2021.11.2-py3-none-any.whl (178 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m178.9/178.9 kB[0m [31m748.6 kB/s[0m eta [36m0:00:00[0ma [36m0:00:01[0m
[?25hRequirement already satisfied: packaging>=20.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from scikit-image->-r PaddleOCR/requirements.txt (line 2)) (21.3)
Requirement already satisfied: networkx>=2.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from scikit-image->-r PaddleOCR/requirements.txt (line 2)) (2.4)
Collecting PyWavelets>=1.1.1
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/ae/56/4441877073d8a5266dbf7b04c7f3dc66f1149c8efb9323e0ef987a9bb1ce/PyWavelets-1.3.0-cp37-cp37m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl (6.4 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m6.4/6.4 MB[0m [31m7.5 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hRequirement already satisfied: pandas in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->-r PaddleOCR/requirements.txt (line 8)) (1.1.5)
Requirement already satisfied: Flask-Babel>=1.0.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->-r PaddleOCR/requirements.txt (line 8)) (1.0.0)
Requirement already satisfied: flask>=1.1.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->-r PaddleOCR/requirements.txt (line 8)) (1.1.1)
Requirement already satisfied: bce-python-sdk in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->-r PaddleOCR/requirements.txt (line 8)) (0.8.53)
Requirement already satisfied: requests in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->-r PaddleOCR/requirements.txt (line 8)) (2.24.0)
Requirement already satisfied: protobuf>=3.11.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->-r PaddleOCR/requirements.txt (line 8)) (3.20.0)
Collecting jarowinkler<2.0.0,>=1.2.0
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/ed/a8/094e87e92bf505dd808ee2652fbcdf73e2599e32de32d24703ecccdba32c/jarowinkler-1.2.1-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (104 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m104.9/104.9 kB[0m [31m6.5 MB/s[0m eta [36m0:00:00[0m
[?25hRequirement already satisfied: cachetools in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from premailer->-r PaddleOCR/requirements.txt (line 13)) (4.0.0)
Collecting cssutils
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/28/4e/6c6d07b1ab43f7b8b939406b562c652c651a66699b8a01fc6b774cbf69de/cssutils-2.6.0-py3-none-any.whl (399 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m399.7/399.7 kB[0m [31m5.3 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hCollecting cssselect
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/3b/d4/3b5c17f00cce85b9a1e6f91096e1cc8e8ede2e1be8e96b87ce1ed09e92c5/cssselect-1.1.0-py2.py3-none-any.whl (16 kB)
Requirement already satisfied: et-xmlfile in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from openpyxl->-r PaddleOCR/requirements.txt (line 14)) (1.0.1)
Requirement already satisfied: jdcal in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from openpyxl->-r PaddleOCR/requirements.txt (line 14)) (1.4.1)
Requirement already satisfied: Werkzeug>=0.15 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from flask>=1.1.1->visualdl->-r PaddleOCR/requirements.txt (line 8)) (0.16.0)
Requirement already satisfied: Jinja2>=2.10.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from flask>=1.1.1->visualdl->-r PaddleOCR/requirements.txt (line 8)) (3.0.0)
Requirement already satisfied: click>=5.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from flask>=1.1.1->visualdl->-r PaddleOCR/requirements.txt (line 8)) (7.0)
Requirement already satisfied: itsdangerous>=0.24 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from flask>=1.1.1->visualdl->-r PaddleOCR/requirements.txt (line 8)) (1.1.0)
Requirement already satisfied: pytz in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask-Babel>=1.0.0->visualdl->-r PaddleOCR/requirements.txt (line 8)) (2019.3)
Requirement already satisfied: Babel>=2.3 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask-Babel>=1.0.0->visualdl->-r PaddleOCR/requirements.txt (line 8)) (2.8.0)
Requirement already satisfied: decorator>=4.3.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from networkx>=2.2->scikit-image->-r PaddleOCR/requirements.txt (line 2)) (4.4.2)
Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from packaging>=20.0->scikit-image->-r PaddleOCR/requirements.txt (line 2)) (3.0.9)
Requirement already satisfied: pycryptodome>=3.8.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from bce-python-sdk->visualdl->-r PaddleOCR/requirements.txt (line 8)) (3.9.9)
Requirement already satisfied: future>=0.6.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from bce-python-sdk->visualdl->-r PaddleOCR/requirements.txt (line 8)) (0.18.0)
Requirement already satisfied: importlib-metadata in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from cssutils->premailer->-r PaddleOCR/requirements.txt (line 13)) (4.2.0)
Requirement already satisfied: python-dateutil>=2.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from matplotlib->imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (2.8.2)
Requirement already satisfied: kiwisolver>=1.0.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from matplotlib->imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (1.1.0)
Requirement already satisfied: cycler>=0.10 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from matplotlib->imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (0.10.0)
Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests->visualdl->-r PaddleOCR/requirements.txt (line 8)) (2019.9.11)
Requirement already satisfied: idna<3,>=2.5 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests->visualdl->-r PaddleOCR/requirements.txt (line 8)) (2.8)
Requirement already satisfied: chardet<4,>=3.0.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests->visualdl->-r PaddleOCR/requirements.txt (line 8)) (3.0.4)
Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests->visualdl->-r PaddleOCR/requirements.txt (line 8)) (1.25.6)
Requirement already satisfied: MarkupSafe>=2.0.0rc2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Jinja2>=2.10.1->flask>=1.1.1->visualdl->-r PaddleOCR/requirements.txt (line 8)) (2.0.1)
Requirement already satisfied: setuptools in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from kiwisolver>=1.0.1->matplotlib->imgaug==0.4.0->-r PaddleOCR/requirements.txt (line 3)) (56.2.0)
Requirement already satisfied: typing-extensions>=3.6.4 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from importlib-metadata->cssutils->premailer->-r PaddleOCR/requirements.txt (line 13)) (4.3.0)
Requirement already satisfied: zipp>=0.5 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from importlib-metadata->cssutils->premailer->-r PaddleOCR/requirements.txt (line 13)) (3.8.1)
Installing collected packages: pyclipper, lmdb, tifffile, shapely, PyWavelets, opencv-contrib-python, lxml, jarowinkler, cssselect, attrdict, scikit-image, rapidfuzz, cssutils, premailer, imgaug
Successfully installed PyWavelets-1.3.0 attrdict-2.0.1 cssselect-1.1.0 cssutils-2.6.0 imgaug-0.4.0 jarowinkler-1.2.1 lmdb-1.3.0 lxml-4.9.1 opencv-contrib-python-4.4.0.46 premailer-3.10.0 pyclipper-1.3.0.post3 rapidfuzz-2.6.0 scikit-image-0.19.3 shapely-1.8.4 tifffile-2021.11.2
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Collecting paddle2onnx
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/be/62/bd17eeccd7cfd3601ae8607371673046158d67f48d095c3055edac2e5250/paddle2onnx-0.9.8-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (2.9 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m2.9/2.9 MB[0m [31m6.6 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hInstalling collected packages: paddle2onnx
Successfully installed paddle2onnx-0.9.8
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Collecting onnx
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/bf/c5/e8edd9bc58192ef964270e2f4600a02cd5e5d0958b81f7abe2ee0a604478/onnx-1.12.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (13.1 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m13.1/13.1 MB[0m [31m10.2 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hRequirement already satisfied: protobuf<=3.20.1,>=3.12.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from onnx) (3.20.0)
Requirement already satisfied: numpy>=1.16.6 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from onnx) (1.19.5)
Requirement already satisfied: typing-extensions>=3.6.2.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from onnx) (4.3.0)
Installing collected packages: onnx
Successfully installed onnx-1.12.0
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Collecting onnxruntime-gpu
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/8b/05/922f716ec6239cb126d26b64ae1c60a91bdf2b27c56ac62ac9199015e489/onnxruntime_gpu-1.12.1-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (111.0 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m111.0/111.0 MB[0m [31m3.8 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hCollecting sympy
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/d0/04/66be21ceb305c66a4b326b0ae44cc4f027a43bc08cac204b48fb45bb3653/sympy-1.10.1-py3-none-any.whl (6.4 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m6.4/6.4 MB[0m [31m8.5 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0mm
[?25hRequirement already satisfied: protobuf in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from onnxruntime-gpu) (3.20.0)
Collecting coloredlogs
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/a7/06/3d6badcf13db419e25b07041d9c7b4a2c331d3f4e7134445ec5df57714cd/coloredlogs-15.0.1-py2.py3-none-any.whl (46 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m46.0/46.0 kB[0m [31m9.3 MB/s[0m eta [36m0:00:00[0m
[?25hRequirement already satisfied: packaging in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from onnxruntime-gpu) (21.3)
Collecting flatbuffers
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/d7/0d/b5bfb553a6ac66d6ec2b6d7f1e814a908fba7188356ac94bb36ae3d905c3/flatbuffers-2.0.7-py2.py3-none-any.whl (26 kB)
Collecting numpy>=1.21.0
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/6d/ad/ff3b21ebfe79a4d25b4a4f8e5cf9fd44a204adb6b33c09010f566f51027a/numpy-1.21.6-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (15.7 MB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m15.7/15.7 MB[0m [31m7.6 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hCollecting humanfriendly>=9.1
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/f0/0f/310fb31e39e2d734ccaa2c0fb981ee41f7bd5056ce9bc29b2248bd569169/humanfriendly-10.0-py2.py3-none-any.whl (86 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m86.8/86.8 kB[0m [31m8.0 MB/s[0m eta [36m0:00:00[0m
[?25hRequirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from packaging->onnxruntime-gpu) (3.0.9)
Collecting mpmath>=0.19
  Downloading https://pypi.tuna.tsinghua.edu.cn/packages/d4/cf/3965bddbb4f1a61c49aacae0e78fd1fe36b5dc36c797b31f30cf07dcbbb7/mpmath-1.2.1-py3-none-any.whl (532 kB)
[2K     [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m532.6/532.6 kB[0m [31m8.4 MB/s[0m eta [36m0:00:00[0ma [36m0:00:01[0m
[?25hInstalling collected packages: mpmath, flatbuffers, sympy, numpy, humanfriendly, coloredlogs, onnxruntime-gpu
  Attempting uninstall: numpy
    Found existing installation: numpy 1.19.5
    Uninstalling numpy-1.19.5:
      Successfully uninstalled numpy-1.19.5
[31mERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
parl 1.4.1 requires pyzmq==18.1.1, but you have pyzmq 23.2.0 which is incompatible.
paddlefsl 1.0.0 requires numpy~=1.19.2, but you have numpy 1.21.6 which is incompatible.[0m[31m
[0mSuccessfully installed coloredlogs-15.0.1 flatbuffers-2.0.7 humanfriendly-10.0 mpmath-1.2.1 numpy-1.21.6 onnxruntime-gpu-1.12.1 sympy-1.10.1

!cp /home/aistudio/work/export_model.py /home/aistudio/PaddleOCR/tools/export_model.py
!cp /home/aistudio/work/rec_svtrnet.yml /home/aistudio/PaddleOCR/configs/rec/rec_svtrnet.yml
!cp /home/aistudio/work/rec_svtrnet.py /home/aistudio/PaddleOCR/ppocr/modeling/backbones/rec_svtrnet.py
!cp /home/aistudio/work/rec_img_aug.py /home/aistudio/PaddleOCR/ppocr/data/imaug/rec_img_aug.py
!cp /home/aistudio/work/en_PP-OCRv3_rec.yml /home/aistudio/PaddleOCR/configs/rec/PP-OCRv3/en_PP-OCRv3_rec.yml

3.1 数据处理

解压数据
处理数据并生成字典及训练集

%cd ~
!unzip -o data/data62843/test_images.zip -d /home/aistudio/data/
!unzip -o data/data62842/train_images.zip -d /home/aistudio/data/
!cp -r data/data62842/train_label.csv /home/aistudio/data/

  inflating: /home/aistudio/data/train_images/13631.jpg

%cd ~

import os 
import random
from zhtools.langconv import Converter
import pandas as pd 


word_list = []
datas = []

converter = Converter('zh-hans')

def is_chinese(uchar):
    """判断一个unicode是否是汉字"""
    if uchar >= u'\u4e00' and uchar<=u'\u9fa5':
        return True
    else:
        return False

def is_number(uchar):
    """判断一个unicode是否是半角数字"""
    if uchar >= u'\u0030' and uchar<=u'\u0039':
        return True
    else:
        return False

def is_english(uchar):
    """判断一个unicode是否是汉字"""
    if uchar >= u'\u0061' and uchar<=u'\u007a':
        return True
    else:
        return False

def Q2B(uchar):
    """单个字符 全角转半角"""
    inside_code = ord(uchar)
    if inside_code == 0x3000:
        inside_code = 0x0020
    else:
        inside_code -= 0xfee0
    if inside_code < 0x0020 or inside_code > 0x7e: #转完之后不是半角字符返回原来的字符
        return uchar
    return chr(inside_code)


# 读取标注文件
data = pd.read_csv('data/train_label.csv',encoding = 'gb2312')
for i in range(len(data)):
    name, label = data.iloc[i,:]
    label = label.replace('　','')
    label = converter.convert(label)
    label.lower()
    new_label = []
    for word in label:
        word = Q2B(word)
        if is_chinese(word) or is_number(word) or is_english(word):
            new_label.append(word)
            if word not in word_list:
                word_list.append(word)
    if new_label!=[]:
        datas.append('%s\t%s\n' % (os.path.join('train_images',name), ''.join(new_label)))

word_list.sort()

# 生成词表
with open('data/vocab.txt', 'w', encoding='UTF-8') as f:
    for word in word_list:
        f.write(word+'\n')

random.shuffle(datas)
# 训练数据95% 验证数据%5
split_num = int(len(datas)*0.95)

print('训练数据：',split_num,'验证数据：',int(len(datas)*0.05))
# 分割数据为训练和验证集
with open('data/train.txt', 'w', encoding='UTF-8') as f:
    for line in datas[:split_num]:
        f.write(line)

with open('data/dev.txt', 'w', encoding='UTF-8') as f:
    for line in datas[split_num:]:
        f.write(line)

/home/aistudio
训练数据： 43520 验证数据： 2290

3.2 模型训练

!cp /home/aistudio/work/en_PP-OCRv3_rec.yml /home/aistudio/PaddleOCR/configs/rec/PP-OCRv3/en_PP-OCRv3_rec.yml

%cd ~
!python3 PaddleOCR/tools/train.py -c PaddleOCR/configs/rec/PP-OCRv3/en_PP-OCRv3_rec.yml -o Global.checkpoints=/home/aistudio/output/rec/PPOCRV3_0.5/best_accuracy

/home/aistudio
[2022/08/24 23:49:37] ppocr INFO: Architecture : 
[2022/08/24 23:49:37] ppocr INFO:     Backbone : 
[2022/08/24 23:49:37] ppocr INFO:         layers : 34
[2022/08/24 23:49:37] ppocr INFO:         name : ResNet
[2022/08/24 23:49:37] ppocr INFO:     Head : 
[2022/08/24 23:49:37] ppocr INFO:         fc_decay : 0
[2022/08/24 23:49:37] ppocr INFO:         name : CTCHead
[2022/08/24 23:49:37] ppocr INFO:     Neck : 
[2022/08/24 23:49:37] ppocr INFO:         encoder_type : rnn
[2022/08/24 23:49:37] ppocr INFO:         hidden_size : 256
[2022/08/24 23:49:37] ppocr INFO:         name : SequenceEncoder
[2022/08/24 23:49:37] ppocr INFO:     Transform : None
[2022/08/24 23:49:37] ppocr INFO:     algorithm : CRNN
[2022/08/24 23:49:37] ppocr INFO:     model_type : rec
[2022/08/24 23:49:37] ppocr INFO: Eval : 
[2022/08/24 23:49:37] ppocr INFO:     dataset : 
[2022/08/24 23:49:37] ppocr INFO:         data_dir : /home/aistudio/data/
[2022/08/24 23:49:37] ppocr INFO:         label_file_list : ['/home/aistudio/data/dev.txt']
[2022/08/24 23:49:37] ppocr INFO:         name : SimpleDataSet
[2022/08/24 23:49:37] ppocr INFO:         transforms : 
[2022/08/24 23:49:37] ppocr INFO:             DecodeImage : 
[2022/08/24 23:49:37] ppocr INFO:                 channel_first : False
[2022/08/24 23:49:37] ppocr INFO:                 img_mode : BGR
[2022/08/24 23:49:37] ppocr INFO:             CTCLabelEncode : None
[2022/08/24 23:49:37] ppocr INFO:             RecResizeImg : 
[2022/08/24 23:49:37] ppocr INFO:                 image_shape : [3, 48, 256]
[2022/08/24 23:49:37] ppocr INFO:             KeepKeys : 
[2022/08/24 23:49:37] ppocr INFO:                 keep_keys : ['image', 'label', 'length']
[2022/08/24 23:49:37] ppocr INFO:     loader : 
[2022/08/24 23:49:37] ppocr INFO:         batch_size_per_card : 128
[2022/08/24 23:49:37] ppocr INFO:         drop_last : False
[2022/08/24 23:49:37] ppocr INFO:         num_workers : 4
[2022/08/24 23:49:37] ppocr INFO:         shuffle : False
[2022/08/24 23:49:37] ppocr INFO: Global : 
[2022/08/24 23:49:37] ppocr INFO:     cal_metric_during_train : True
[2022/08/24 23:49:37] ppocr INFO:     character_dict_path : /home/aistudio/data/vocab.txt
[2022/08/24 23:49:37] ppocr INFO:     checkpoints : /home/aistudio/output/rec/PPOCRV3_0.5/best_accuracy
[2022/08/24 23:49:37] ppocr INFO:     debug : False
[2022/08/24 23:49:37] ppocr INFO:     distributed : False
[2022/08/24 23:49:37] ppocr INFO:     epoch_num : 100
[2022/08/24 23:49:37] ppocr INFO:     eval_batch_step : [0, 2000]
[2022/08/24 23:49:37] ppocr INFO:     infer_img : /home/aistudio/data/test_images/
[2022/08/24 23:49:37] ppocr INFO:     infer_mode : False
[2022/08/24 23:49:37] ppocr INFO:     log_smooth_window : 20
[2022/08/24 23:49:37] ppocr INFO:     max_text_length : 25
[2022/08/24 23:49:37] ppocr INFO:     pretrained_model : None
[2022/08/24 23:49:37] ppocr INFO:     print_batch_step : 20
[2022/08/24 23:49:37] ppocr INFO:     save_epoch_step : 10
[2022/08/24 23:49:37] ppocr INFO:     save_inference_dir : None
[2022/08/24 23:49:37] ppocr INFO:     save_model_dir : ./output/rec/PPOCRV3_0.5/
[2022/08/24 23:49:37] ppocr INFO:     save_res_path : ./output/rec/predicts_ppocrv3_0.5_en.txt
[2022/08/24 23:49:37] ppocr INFO:     use_gpu : True
[2022/08/24 23:49:37] ppocr INFO:     use_space_char : True
[2022/08/24 23:49:37] ppocr INFO:     use_visualdl : False
[2022/08/24 23:49:37] ppocr INFO: Loss : 
[2022/08/24 23:49:37] ppocr INFO:     name : CTCLoss
[2022/08/24 23:49:37] ppocr INFO: Metric : 
[2022/08/24 23:49:37] ppocr INFO:     main_indicator : acc
[2022/08/24 23:49:37] ppocr INFO:     name : RecMetric
[2022/08/24 23:49:37] ppocr INFO: Optimizer : 
[2022/08/24 23:49:37] ppocr INFO:     beta1 : 0.9
[2022/08/24 23:49:37] ppocr INFO:     beta2 : 0.999
[2022/08/24 23:49:37] ppocr INFO:     lr : 
[2022/08/24 23:49:37] ppocr INFO:         learning_rate : 0.001
[2022/08/24 23:49:37] ppocr INFO:         name : Cosine
[2022/08/24 23:49:37] ppocr INFO:         warmup_epoch : 5
[2022/08/24 23:49:37] ppocr INFO:     name : Adam
[2022/08/24 23:49:37] ppocr INFO:     regularizer : 
[2022/08/24 23:49:37] ppocr INFO:         factor : 3e-05
[2022/08/24 23:49:37] ppocr INFO:         name : L2
[2022/08/24 23:49:37] ppocr INFO: PostProcess : 
[2022/08/24 23:49:37] ppocr INFO:     name : CTCLabelDecode
[2022/08/24 23:49:37] ppocr INFO: Train : 
[2022/08/24 23:49:37] ppocr INFO:     dataset : 
[2022/08/24 23:49:37] ppocr INFO:         data_dir : /home/aistudio/data/
[2022/08/24 23:49:37] ppocr INFO:         ext_op_transform_idx : 1
[2022/08/24 23:49:37] ppocr INFO:         label_file_list : ['/home/aistudio/data/train.txt']
[2022/08/24 23:49:37] ppocr INFO:         name : SimpleDataSet
[2022/08/24 23:49:37] ppocr INFO:         transforms : 
[2022/08/24 23:49:37] ppocr INFO:             DecodeImage : 
[2022/08/24 23:49:37] ppocr INFO:                 channel_first : False
[2022/08/24 23:49:37] ppocr INFO:                 img_mode : BGR
[2022/08/24 23:49:37] ppocr INFO:             RecConAug : 
[2022/08/24 23:49:37] ppocr INFO:                 ext_data_num : 2
[2022/08/24 23:49:37] ppocr INFO:                 image_shape : [48, 256, 3]
[2022/08/24 23:49:37] ppocr INFO:                 max_text_length : 25
[2022/08/24 23:49:37] ppocr INFO:                 prob : 0.5
[2022/08/24 23:49:37] ppocr INFO:             RecAug : None
[2022/08/24 23:49:37] ppocr INFO:             CTCLabelEncode : None
[2022/08/24 23:49:37] ppocr INFO:             RecResizeImg : 
[2022/08/24 23:49:37] ppocr INFO:                 image_shape : [3, 48, 256]
[2022/08/24 23:49:37] ppocr INFO:             KeepKeys : 
[2022/08/24 23:49:37] ppocr INFO:                 keep_keys : ['image', 'label', 'length']
[2022/08/24 23:49:37] ppocr INFO:     loader : 
[2022/08/24 23:49:37] ppocr INFO:         batch_size_per_card : 128
[2022/08/24 23:49:37] ppocr INFO:         drop_last : True
[2022/08/24 23:49:37] ppocr INFO:         num_workers : 4
[2022/08/24 23:49:37] ppocr INFO:         shuffle : True
[2022/08/24 23:49:37] ppocr INFO: profiler_options : None
[2022/08/24 23:49:37] ppocr INFO: train with paddle 2.3.1 and device Place(gpu:0)
[2022/08/24 23:49:37] ppocr INFO: Initialize indexs of datasets:['/home/aistudio/data/train.txt']
[2022/08/24 23:49:37] ppocr INFO: Initialize indexs of datasets:['/home/aistudio/data/dev.txt']
W0824 23:49:37.897917  1210 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0824 23:49:37.902611  1210 gpu_resources.cc:91] device: 0, cuDNN Version: 7.6.
[2022/08/24 23:49:39] ppocr INFO: resume from /home/aistudio/output/rec/PPOCRV3_0.5/best_accuracy
[2022/08/24 23:49:39] ppocr INFO: train dataloader has 340 iters
[2022/08/24 23:49:39] ppocr INFO: valid dataloader has 18 iters
[2022/08/24 23:49:39] ppocr INFO: During the training process, after the 0th iteration, an evaluation is run every 2000 iterations
[2022/08/24 23:50:04] ppocr INFO: epoch: [78/100], global_step: 20, lr: 0.000187, acc: 0.921875, norm_edit_dis: 0.976256, loss: 0.848411, avg_reader_cost: 0.08716 s, avg_batch_cost: 1.25074 s, avg_samples: 128.0, ips: 102.33932 samples/s, eta: 2:42:35
[2022/08/24 23:50:25] ppocr INFO: epoch: [78/100], global_step: 40, lr: 0.000187, acc: 0.917969, norm_edit_dis: 0.979142, loss: 0.723838, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.05690 s, avg_samples: 128.0, ips: 121.10871 samples/s, eta: 2:29:36
[2022/08/24 23:50:47] ppocr INFO: epoch: [78/100], global_step: 60, lr: 0.000186, acc: 0.914062, norm_edit_dis: 0.974852, loss: 0.820106, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.05893 s, avg_samples: 128.0, ips: 120.87714 samples/s, eta: 2:25:08
[2022/08/24 23:51:08] ppocr INFO: epoch: [78/100], global_step: 80, lr: 0.000185, acc: 0.925781, norm_edit_dis: 0.979751, loss: 0.637728, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06085 s, avg_samples: 128.0, ips: 120.65772 samples/s, eta: 2:22:47
[2022/08/24 23:51:29] ppocr INFO: epoch: [78/100], global_step: 100, lr: 0.000185, acc: 0.921875, norm_edit_dis: 0.975549, loss: 0.649261, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06341 s, avg_samples: 128.0, ips: 120.36784 samples/s, eta: 2:21:17
[2022/08/24 23:51:50] ppocr INFO: epoch: [78/100], global_step: 120, lr: 0.000184, acc: 0.917969, norm_edit_dis: 0.979270, loss: 0.739498, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06461 s, avg_samples: 128.0, ips: 120.23224 samples/s, eta: 2:20:12
[2022/08/24 23:52:12] ppocr INFO: epoch: [78/100], global_step: 140, lr: 0.000183, acc: 0.917969, norm_edit_dis: 0.980207, loss: 0.690302, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06589 s, avg_samples: 128.0, ips: 120.08711 samples/s, eta: 2:19:21
[2022/08/24 23:52:33] ppocr INFO: epoch: [78/100], global_step: 160, lr: 0.000182, acc: 0.914062, norm_edit_dis: 0.976779, loss: 0.783364, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06516 s, avg_samples: 128.0, ips: 120.16962 samples/s, eta: 2:18:37
[2022/08/24 23:52:54] ppocr INFO: epoch: [78/100], global_step: 180, lr: 0.000182, acc: 0.914062, norm_edit_dis: 0.979177, loss: 0.763737, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06669 s, avg_samples: 128.0, ips: 119.99745 samples/s, eta: 2:17:59
[2022/08/24 23:53:16] ppocr INFO: epoch: [78/100], global_step: 200, lr: 0.000181, acc: 0.910156, norm_edit_dis: 0.976583, loss: 0.784274, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06596 s, avg_samples: 128.0, ips: 120.07902 samples/s, eta: 2:17:24
[2022/08/24 23:53:37] ppocr INFO: epoch: [78/100], global_step: 220, lr: 0.000180, acc: 0.914062, norm_edit_dis: 0.974881, loss: 0.690789, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06569 s, avg_samples: 128.0, ips: 120.10946 samples/s, eta: 2:16:51
[2022/08/24 23:53:58] ppocr INFO: epoch: [78/100], global_step: 240, lr: 0.000180, acc: 0.917969, norm_edit_dis: 0.977017, loss: 0.698131, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06752 s, avg_samples: 128.0, ips: 119.90426 samples/s, eta: 2:16:21
[2022/08/24 23:54:20] ppocr INFO: epoch: [78/100], global_step: 260, lr: 0.000179, acc: 0.933594, norm_edit_dis: 0.978560, loss: 0.669779, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06569 s, avg_samples: 128.0, ips: 120.11024 samples/s, eta: 2:15:52
[2022/08/24 23:54:41] ppocr INFO: epoch: [78/100], global_step: 280, lr: 0.000178, acc: 0.921875, norm_edit_dis: 0.977822, loss: 0.729766, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06517 s, avg_samples: 128.0, ips: 120.16849 samples/s, eta: 2:15:23
[2022/08/24 23:55:02] ppocr INFO: epoch: [78/100], global_step: 300, lr: 0.000177, acc: 0.921875, norm_edit_dis: 0.976196, loss: 0.690487, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06405 s, avg_samples: 128.0, ips: 120.29561 samples/s, eta: 2:14:55
[2022/08/24 23:55:24] ppocr INFO: epoch: [78/100], global_step: 320, lr: 0.000177, acc: 0.917969, norm_edit_dis: 0.977399, loss: 0.707437, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06427 s, avg_samples: 128.0, ips: 120.27019 samples/s, eta: 2:14:27
[2022/08/24 23:55:45] ppocr INFO: epoch: [78/100], global_step: 340, lr: 0.000176, acc: 0.925781, norm_edit_dis: 0.980298, loss: 0.704222, avg_reader_cost: 0.00050 s, avg_batch_cost: 1.06513 s, avg_samples: 128.0, ips: 120.17286 samples/s, eta: 2:14:01
[2022/08/24 23:55:48] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/24 23:56:11] ppocr INFO: epoch: [79/100], global_step: 360, lr: 0.000175, acc: 0.921875, norm_edit_dis: 0.979195, loss: 0.590463, avg_reader_cost: 0.21898 s, avg_batch_cost: 1.28838 s, avg_samples: 128.0, ips: 99.34933 samples/s, eta: 2:15:08
[2022/08/24 23:56:32] ppocr INFO: epoch: [79/100], global_step: 380, lr: 0.000175, acc: 0.921875, norm_edit_dis: 0.982235, loss: 0.655956, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06468 s, avg_samples: 128.0, ips: 120.22422 samples/s, eta: 2:14:38
[2022/08/24 23:56:53] ppocr INFO: epoch: [79/100], global_step: 400, lr: 0.000174, acc: 0.929687, norm_edit_dis: 0.981516, loss: 0.487102, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06126 s, avg_samples: 128.0, ips: 120.61107 samples/s, eta: 2:14:07
[2022/08/24 23:57:14] ppocr INFO: epoch: [79/100], global_step: 420, lr: 0.000173, acc: 0.921875, norm_edit_dis: 0.978573, loss: 0.580915, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06147 s, avg_samples: 128.0, ips: 120.58743 samples/s, eta: 2:13:37
[2022/08/24 23:57:36] ppocr INFO: epoch: [79/100], global_step: 440, lr: 0.000172, acc: 0.925781, norm_edit_dis: 0.980269, loss: 0.648877, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06137 s, avg_samples: 128.0, ips: 120.59934 samples/s, eta: 2:13:08
[2022/08/24 23:57:57] ppocr INFO: epoch: [79/100], global_step: 460, lr: 0.000172, acc: 0.921875, norm_edit_dis: 0.980715, loss: 0.554099, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06260 s, avg_samples: 128.0, ips: 120.45872 samples/s, eta: 2:12:40
[2022/08/24 23:58:18] ppocr INFO: epoch: [79/100], global_step: 480, lr: 0.000171, acc: 0.929687, norm_edit_dis: 0.979484, loss: 0.524505, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06380 s, avg_samples: 128.0, ips: 120.32341 samples/s, eta: 2:12:13
[2022/08/24 23:58:40] ppocr INFO: epoch: [79/100], global_step: 500, lr: 0.000170, acc: 0.921875, norm_edit_dis: 0.982501, loss: 0.533287, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06556 s, avg_samples: 128.0, ips: 120.12438 samples/s, eta: 2:11:47
[2022/08/24 23:59:01] ppocr INFO: epoch: [79/100], global_step: 520, lr: 0.000170, acc: 0.933594, norm_edit_dis: 0.978490, loss: 0.530778, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06658 s, avg_samples: 128.0, ips: 120.00925 samples/s, eta: 2:11:21
[2022/08/24 23:59:22] ppocr INFO: epoch: [79/100], global_step: 540, lr: 0.000169, acc: 0.921875, norm_edit_dis: 0.981197, loss: 0.730163, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06672 s, avg_samples: 128.0, ips: 119.99385 samples/s, eta: 2:10:56
[2022/08/24 23:59:44] ppocr INFO: epoch: [79/100], global_step: 560, lr: 0.000168, acc: 0.917969, norm_edit_dis: 0.977392, loss: 0.713846, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06635 s, avg_samples: 128.0, ips: 120.03589 samples/s, eta: 2:10:31
[2022/08/25 00:00:05] ppocr INFO: epoch: [79/100], global_step: 580, lr: 0.000168, acc: 0.937500, norm_edit_dis: 0.981308, loss: 0.630154, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06565 s, avg_samples: 128.0, ips: 120.11402 samples/s, eta: 2:10:06
[2022/08/25 00:00:26] ppocr INFO: epoch: [79/100], global_step: 600, lr: 0.000167, acc: 0.925781, norm_edit_dis: 0.983745, loss: 0.612132, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06523 s, avg_samples: 128.0, ips: 120.16203 samples/s, eta: 2:09:42
[2022/08/25 00:00:48] ppocr INFO: epoch: [79/100], global_step: 620, lr: 0.000166, acc: 0.929687, norm_edit_dis: 0.985034, loss: 0.551495, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06590 s, avg_samples: 128.0, ips: 120.08630 samples/s, eta: 2:09:17
[2022/08/25 00:01:09] ppocr INFO: epoch: [79/100], global_step: 640, lr: 0.000166, acc: 0.929687, norm_edit_dis: 0.981490, loss: 0.504669, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06508 s, avg_samples: 128.0, ips: 120.17863 samples/s, eta: 2:08:53
[2022/08/25 00:01:30] ppocr INFO: epoch: [79/100], global_step: 660, lr: 0.000165, acc: 0.929687, norm_edit_dis: 0.977512, loss: 0.479940, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06574 s, avg_samples: 128.0, ips: 120.10470 samples/s, eta: 2:08:29
[2022/08/25 00:01:52] ppocr INFO: epoch: [79/100], global_step: 680, lr: 0.000164, acc: 0.925781, norm_edit_dis: 0.980683, loss: 0.530971, avg_reader_cost: 0.00075 s, avg_batch_cost: 1.06762 s, avg_samples: 128.0, ips: 119.89323 samples/s, eta: 2:08:06
[2022/08/25 00:01:54] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:02:17] ppocr INFO: epoch: [80/100], global_step: 700, lr: 0.000164, acc: 0.929687, norm_edit_dis: 0.981334, loss: 0.553367, avg_reader_cost: 0.22561 s, avg_batch_cost: 1.29311 s, avg_samples: 128.0, ips: 98.98584 samples/s, eta: 2:08:28
[2022/08/25 00:02:39] ppocr INFO: epoch: [80/100], global_step: 720, lr: 0.000163, acc: 0.921875, norm_edit_dis: 0.982988, loss: 0.607721, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06371 s, avg_samples: 128.0, ips: 120.33311 samples/s, eta: 2:08:03
[2022/08/25 00:03:00] ppocr INFO: epoch: [80/100], global_step: 740, lr: 0.000162, acc: 0.937500, norm_edit_dis: 0.985107, loss: 0.474372, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06162 s, avg_samples: 128.0, ips: 120.57025 samples/s, eta: 2:07:37
[2022/08/25 00:03:21] ppocr INFO: epoch: [80/100], global_step: 760, lr: 0.000161, acc: 0.937500, norm_edit_dis: 0.987567, loss: 0.508198, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06101 s, avg_samples: 128.0, ips: 120.63993 samples/s, eta: 2:07:12
[2022/08/25 00:03:42] ppocr INFO: epoch: [80/100], global_step: 780, lr: 0.000161, acc: 0.933594, norm_edit_dis: 0.983355, loss: 0.522368, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06215 s, avg_samples: 128.0, ips: 120.50999 samples/s, eta: 2:06:47
[2022/08/25 00:04:04] ppocr INFO: epoch: [80/100], global_step: 800, lr: 0.000160, acc: 0.937500, norm_edit_dis: 0.982000, loss: 0.627857, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06390 s, avg_samples: 128.0, ips: 120.31220 samples/s, eta: 2:06:22
[2022/08/25 00:04:25] ppocr INFO: epoch: [80/100], global_step: 820, lr: 0.000159, acc: 0.937500, norm_edit_dis: 0.985607, loss: 0.414067, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06434 s, avg_samples: 128.0, ips: 120.26250 samples/s, eta: 2:05:58
[2022/08/25 00:04:46] ppocr INFO: epoch: [80/100], global_step: 840, lr: 0.000159, acc: 0.925781, norm_edit_dis: 0.982685, loss: 0.593329, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06572 s, avg_samples: 128.0, ips: 120.10686 samples/s, eta: 2:05:34
[2022/08/25 00:05:08] ppocr INFO: epoch: [80/100], global_step: 860, lr: 0.000158, acc: 0.925781, norm_edit_dis: 0.984307, loss: 0.493968, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06575 s, avg_samples: 128.0, ips: 120.10283 samples/s, eta: 2:05:10
[2022/08/25 00:05:29] ppocr INFO: epoch: [80/100], global_step: 880, lr: 0.000157, acc: 0.929687, norm_edit_dis: 0.985507, loss: 0.611084, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06771 s, avg_samples: 128.0, ips: 119.88317 samples/s, eta: 2:04:47
[2022/08/25 00:05:50] ppocr INFO: epoch: [80/100], global_step: 900, lr: 0.000157, acc: 0.925781, norm_edit_dis: 0.986168, loss: 0.550814, avg_reader_cost: 0.00047 s, avg_batch_cost: 1.06819 s, avg_samples: 128.0, ips: 119.82892 samples/s, eta: 2:04:24
[2022/08/25 00:06:12] ppocr INFO: epoch: [80/100], global_step: 920, lr: 0.000156, acc: 0.929687, norm_edit_dis: 0.979733, loss: 0.654225, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06684 s, avg_samples: 128.0, ips: 119.98056 samples/s, eta: 2:04:00
[2022/08/25 00:06:33] ppocr INFO: epoch: [80/100], global_step: 940, lr: 0.000155, acc: 0.929687, norm_edit_dis: 0.979289, loss: 0.634137, avg_reader_cost: 0.00058 s, avg_batch_cost: 1.06720 s, avg_samples: 128.0, ips: 119.93967 samples/s, eta: 2:03:37
[2022/08/25 00:06:54] ppocr INFO: epoch: [80/100], global_step: 960, lr: 0.000155, acc: 0.933594, norm_edit_dis: 0.984068, loss: 0.567261, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06473 s, avg_samples: 128.0, ips: 120.21781 samples/s, eta: 2:03:14
[2022/08/25 00:07:16] ppocr INFO: epoch: [80/100], global_step: 980, lr: 0.000154, acc: 0.929687, norm_edit_dis: 0.985962, loss: 0.525866, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06493 s, avg_samples: 128.0, ips: 120.19619 samples/s, eta: 2:02:50
[2022/08/25 00:07:37] ppocr INFO: epoch: [80/100], global_step: 1000, lr: 0.000153, acc: 0.929687, norm_edit_dis: 0.980172, loss: 0.649229, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06608 s, avg_samples: 128.0, ips: 120.06549 samples/s, eta: 2:02:27
[2022/08/25 00:07:58] ppocr INFO: epoch: [80/100], global_step: 1020, lr: 0.000153, acc: 0.925781, norm_edit_dis: 0.982380, loss: 0.590779, avg_reader_cost: 0.00049 s, avg_batch_cost: 1.06533 s, avg_samples: 128.0, ips: 120.15004 samples/s, eta: 2:02:04
[2022/08/25 00:08:01] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:08:04] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/iter_epoch_80
[2022/08/25 00:08:27] ppocr INFO: epoch: [81/100], global_step: 1040, lr: 0.000152, acc: 0.929687, norm_edit_dis: 0.984479, loss: 0.462545, avg_reader_cost: 0.35295 s, avg_batch_cost: 1.41917 s, avg_samples: 128.0, ips: 90.19376 samples/s, eta: 2:02:27
[2022/08/25 00:08:48] ppocr INFO: epoch: [81/100], global_step: 1060, lr: 0.000151, acc: 0.941406, norm_edit_dis: 0.986592, loss: 0.454751, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06428 s, avg_samples: 128.0, ips: 120.26861 samples/s, eta: 2:02:03
[2022/08/25 00:09:09] ppocr INFO: epoch: [81/100], global_step: 1080, lr: 0.000151, acc: 0.933594, norm_edit_dis: 0.981600, loss: 0.566702, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06596 s, avg_samples: 128.0, ips: 120.07913 samples/s, eta: 2:01:39
[2022/08/25 00:09:31] ppocr INFO: epoch: [81/100], global_step: 1100, lr: 0.000150, acc: 0.937500, norm_edit_dis: 0.987108, loss: 0.490453, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06643 s, avg_samples: 128.0, ips: 120.02608 samples/s, eta: 2:01:15
[2022/08/25 00:09:52] ppocr INFO: epoch: [81/100], global_step: 1120, lr: 0.000149, acc: 0.941406, norm_edit_dis: 0.984718, loss: 0.466426, avg_reader_cost: 0.00047 s, avg_batch_cost: 1.06468 s, avg_samples: 128.0, ips: 120.22440 samples/s, eta: 2:00:51
[2022/08/25 00:10:13] ppocr INFO: epoch: [81/100], global_step: 1140, lr: 0.000149, acc: 0.945312, norm_edit_dis: 0.985852, loss: 0.433972, avg_reader_cost: 0.00048 s, avg_batch_cost: 1.06306 s, avg_samples: 128.0, ips: 120.40766 samples/s, eta: 2:00:28
[2022/08/25 00:10:35] ppocr INFO: epoch: [81/100], global_step: 1160, lr: 0.000148, acc: 0.937500, norm_edit_dis: 0.982400, loss: 0.543359, avg_reader_cost: 0.00048 s, avg_batch_cost: 1.06510 s, avg_samples: 128.0, ips: 120.17678 samples/s, eta: 2:00:04
[2022/08/25 00:10:56] ppocr INFO: epoch: [81/100], global_step: 1180, lr: 0.000147, acc: 0.929687, norm_edit_dis: 0.981551, loss: 0.540491, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06615 s, avg_samples: 128.0, ips: 120.05852 samples/s, eta: 1:59:41
[2022/08/25 00:11:17] ppocr INFO: epoch: [81/100], global_step: 1200, lr: 0.000147, acc: 0.921875, norm_edit_dis: 0.982034, loss: 0.489591, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06705 s, avg_samples: 128.0, ips: 119.95667 samples/s, eta: 1:59:17
[2022/08/25 00:11:39] ppocr INFO: epoch: [81/100], global_step: 1220, lr: 0.000146, acc: 0.937500, norm_edit_dis: 0.986228, loss: 0.567585, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06855 s, avg_samples: 128.0, ips: 119.78900 samples/s, eta: 1:58:54
[2022/08/25 00:12:00] ppocr INFO: epoch: [81/100], global_step: 1240, lr: 0.000145, acc: 0.949219, norm_edit_dis: 0.988662, loss: 0.510602, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06711 s, avg_samples: 128.0, ips: 119.94982 samples/s, eta: 1:58:31
[2022/08/25 00:12:21] ppocr INFO: epoch: [81/100], global_step: 1260, lr: 0.000145, acc: 0.937500, norm_edit_dis: 0.984034, loss: 0.486967, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06657 s, avg_samples: 128.0, ips: 120.01062 samples/s, eta: 1:58:08
[2022/08/25 00:12:43] ppocr INFO: epoch: [81/100], global_step: 1280, lr: 0.000144, acc: 0.937500, norm_edit_dis: 0.983651, loss: 0.466637, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06852 s, avg_samples: 128.0, ips: 119.79222 samples/s, eta: 1:57:45
[2022/08/25 00:13:04] ppocr INFO: epoch: [81/100], global_step: 1300, lr: 0.000144, acc: 0.937500, norm_edit_dis: 0.984569, loss: 0.537786, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06798 s, avg_samples: 128.0, ips: 119.85213 samples/s, eta: 1:57:22
[2022/08/25 00:13:25] ppocr INFO: epoch: [81/100], global_step: 1320, lr: 0.000143, acc: 0.921875, norm_edit_dis: 0.980735, loss: 0.484260, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06644 s, avg_samples: 128.0, ips: 120.02499 samples/s, eta: 1:57:00
[2022/08/25 00:13:47] ppocr INFO: epoch: [81/100], global_step: 1340, lr: 0.000142, acc: 0.929687, norm_edit_dis: 0.986103, loss: 0.587702, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06553 s, avg_samples: 128.0, ips: 120.12846 samples/s, eta: 1:56:37
[2022/08/25 00:14:08] ppocr INFO: epoch: [81/100], global_step: 1360, lr: 0.000142, acc: 0.941406, norm_edit_dis: 0.985967, loss: 0.482302, avg_reader_cost: 0.00050 s, avg_batch_cost: 1.06487 s, avg_samples: 128.0, ips: 120.20204 samples/s, eta: 1:56:13
[2022/08/25 00:14:11] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:14:34] ppocr INFO: epoch: [82/100], global_step: 1380, lr: 0.000141, acc: 0.941406, norm_edit_dis: 0.985963, loss: 0.506070, avg_reader_cost: 0.23071 s, avg_batch_cost: 1.29817 s, avg_samples: 128.0, ips: 98.60047 samples/s, eta: 1:56:12
[2022/08/25 00:14:55] ppocr INFO: epoch: [82/100], global_step: 1400, lr: 0.000140, acc: 0.945312, norm_edit_dis: 0.988737, loss: 0.497095, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06517 s, avg_samples: 128.0, ips: 120.16851 samples/s, eta: 1:55:49
[2022/08/25 00:15:17] ppocr INFO: epoch: [82/100], global_step: 1420, lr: 0.000140, acc: 0.941406, norm_edit_dis: 0.986869, loss: 0.512465, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06544 s, avg_samples: 128.0, ips: 120.13780 samples/s, eta: 1:55:26
[2022/08/25 00:15:38] ppocr INFO: epoch: [82/100], global_step: 1440, lr: 0.000139, acc: 0.933594, norm_edit_dis: 0.982531, loss: 0.496225, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06436 s, avg_samples: 128.0, ips: 120.26047 samples/s, eta: 1:55:03
[2022/08/25 00:15:59] ppocr INFO: epoch: [82/100], global_step: 1460, lr: 0.000138, acc: 0.937500, norm_edit_dis: 0.986396, loss: 0.458809, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06487 s, avg_samples: 128.0, ips: 120.20276 samples/s, eta: 1:54:39
[2022/08/25 00:16:21] ppocr INFO: epoch: [82/100], global_step: 1480, lr: 0.000138, acc: 0.937500, norm_edit_dis: 0.987730, loss: 0.457735, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06719 s, avg_samples: 128.0, ips: 119.94078 samples/s, eta: 1:54:17
[2022/08/25 00:16:42] ppocr INFO: epoch: [82/100], global_step: 1500, lr: 0.000137, acc: 0.929687, norm_edit_dis: 0.985718, loss: 0.483482, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06678 s, avg_samples: 128.0, ips: 119.98718 samples/s, eta: 1:53:54
[2022/08/25 00:17:03] ppocr INFO: epoch: [82/100], global_step: 1520, lr: 0.000136, acc: 0.925781, norm_edit_dis: 0.982923, loss: 0.501096, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06530 s, avg_samples: 128.0, ips: 120.15401 samples/s, eta: 1:53:31
[2022/08/25 00:17:25] ppocr INFO: epoch: [82/100], global_step: 1540, lr: 0.000136, acc: 0.949219, norm_edit_dis: 0.986432, loss: 0.375381, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06655 s, avg_samples: 128.0, ips: 120.01331 samples/s, eta: 1:53:08
[2022/08/25 00:17:46] ppocr INFO: epoch: [82/100], global_step: 1560, lr: 0.000135, acc: 0.929687, norm_edit_dis: 0.984012, loss: 0.516874, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06620 s, avg_samples: 128.0, ips: 120.05267 samples/s, eta: 1:52:45
[2022/08/25 00:18:07] ppocr INFO: epoch: [82/100], global_step: 1580, lr: 0.000135, acc: 0.945312, norm_edit_dis: 0.988512, loss: 0.360305, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06704 s, avg_samples: 128.0, ips: 119.95829 samples/s, eta: 1:52:22
[2022/08/25 00:18:29] ppocr INFO: epoch: [82/100], global_step: 1600, lr: 0.000134, acc: 0.945312, norm_edit_dis: 0.985832, loss: 0.447563, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06894 s, avg_samples: 128.0, ips: 119.74501 samples/s, eta: 1:52:00
[2022/08/25 00:18:50] ppocr INFO: epoch: [82/100], global_step: 1620, lr: 0.000133, acc: 0.953125, norm_edit_dis: 0.987288, loss: 0.411874, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06746 s, avg_samples: 128.0, ips: 119.91112 samples/s, eta: 1:51:37
[2022/08/25 00:19:11] ppocr INFO: epoch: [82/100], global_step: 1640, lr: 0.000133, acc: 0.937500, norm_edit_dis: 0.986980, loss: 0.427150, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06770 s, avg_samples: 128.0, ips: 119.88331 samples/s, eta: 1:51:15
[2022/08/25 00:19:33] ppocr INFO: epoch: [82/100], global_step: 1660, lr: 0.000132, acc: 0.945312, norm_edit_dis: 0.987128, loss: 0.351825, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06605 s, avg_samples: 128.0, ips: 120.06988 samples/s, eta: 1:50:52
[2022/08/25 00:19:54] ppocr INFO: epoch: [82/100], global_step: 1680, lr: 0.000131, acc: 0.949219, norm_edit_dis: 0.988312, loss: 0.389657, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06428 s, avg_samples: 128.0, ips: 120.26924 samples/s, eta: 1:50:29
[2022/08/25 00:20:15] ppocr INFO: epoch: [82/100], global_step: 1700, lr: 0.000131, acc: 0.941406, norm_edit_dis: 0.986905, loss: 0.347384, avg_reader_cost: 0.00075 s, avg_batch_cost: 1.06623 s, avg_samples: 128.0, ips: 120.04934 samples/s, eta: 1:50:07
[2022/08/25 00:20:18] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:20:41] ppocr INFO: epoch: [83/100], global_step: 1720, lr: 0.000130, acc: 0.945312, norm_edit_dis: 0.985691, loss: 0.396339, avg_reader_cost: 0.22177 s, avg_batch_cost: 1.29017 s, avg_samples: 128.0, ips: 99.21141 samples/s, eta: 1:50:00
[2022/08/25 00:21:02] ppocr INFO: epoch: [83/100], global_step: 1740, lr: 0.000130, acc: 0.949219, norm_edit_dis: 0.986933, loss: 0.416281, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06239 s, avg_samples: 128.0, ips: 120.48252 samples/s, eta: 1:49:37
[2022/08/25 00:21:24] ppocr INFO: epoch: [83/100], global_step: 1760, lr: 0.000129, acc: 0.941406, norm_edit_dis: 0.986900, loss: 0.443401, avg_reader_cost: 0.00056 s, avg_batch_cost: 1.06221 s, avg_samples: 128.0, ips: 120.50329 samples/s, eta: 1:49:14
[2022/08/25 00:21:45] ppocr INFO: epoch: [83/100], global_step: 1780, lr: 0.000128, acc: 0.941406, norm_edit_dis: 0.986221, loss: 0.381654, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06462 s, avg_samples: 128.0, ips: 120.23058 samples/s, eta: 1:48:51
[2022/08/25 00:22:06] ppocr INFO: epoch: [83/100], global_step: 1800, lr: 0.000128, acc: 0.945312, norm_edit_dis: 0.985342, loss: 0.413196, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06545 s, avg_samples: 128.0, ips: 120.13708 samples/s, eta: 1:48:29
[2022/08/25 00:22:28] ppocr INFO: epoch: [83/100], global_step: 1820, lr: 0.000127, acc: 0.945312, norm_edit_dis: 0.989123, loss: 0.379660, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06559 s, avg_samples: 128.0, ips: 120.12134 samples/s, eta: 1:48:06
[2022/08/25 00:22:49] ppocr INFO: epoch: [83/100], global_step: 1840, lr: 0.000126, acc: 0.937500, norm_edit_dis: 0.984251, loss: 0.558575, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06413 s, avg_samples: 128.0, ips: 120.28552 samples/s, eta: 1:47:43
[2022/08/25 00:23:10] ppocr INFO: epoch: [83/100], global_step: 1860, lr: 0.000126, acc: 0.953125, norm_edit_dis: 0.990365, loss: 0.294455, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06317 s, avg_samples: 128.0, ips: 120.39414 samples/s, eta: 1:47:21
[2022/08/25 00:23:32] ppocr INFO: epoch: [83/100], global_step: 1880, lr: 0.000125, acc: 0.953125, norm_edit_dis: 0.988034, loss: 0.486319, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06541 s, avg_samples: 128.0, ips: 120.14144 samples/s, eta: 1:46:58
[2022/08/25 00:23:53] ppocr INFO: epoch: [83/100], global_step: 1900, lr: 0.000125, acc: 0.949219, norm_edit_dis: 0.988835, loss: 0.420262, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06665 s, avg_samples: 128.0, ips: 120.00213 samples/s, eta: 1:46:35
[2022/08/25 00:24:14] ppocr INFO: epoch: [83/100], global_step: 1920, lr: 0.000124, acc: 0.941406, norm_edit_dis: 0.987317, loss: 0.430825, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06590 s, avg_samples: 128.0, ips: 120.08596 samples/s, eta: 1:46:13
[2022/08/25 00:24:35] ppocr INFO: epoch: [83/100], global_step: 1940, lr: 0.000123, acc: 0.945312, norm_edit_dis: 0.988620, loss: 0.467768, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06418 s, avg_samples: 128.0, ips: 120.28057 samples/s, eta: 1:45:50
[2022/08/25 00:24:57] ppocr INFO: epoch: [83/100], global_step: 1960, lr: 0.000123, acc: 0.949219, norm_edit_dis: 0.986084, loss: 0.393550, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06569 s, avg_samples: 128.0, ips: 120.10976 samples/s, eta: 1:45:28
[2022/08/25 00:25:18] ppocr INFO: epoch: [83/100], global_step: 1980, lr: 0.000122, acc: 0.949219, norm_edit_dis: 0.985512, loss: 0.368335, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06534 s, avg_samples: 128.0, ips: 120.14948 samples/s, eta: 1:45:05
[2022/08/25 00:25:39] ppocr INFO: epoch: [83/100], global_step: 2000, lr: 0.000122, acc: 0.957031, norm_edit_dis: 0.986711, loss: 0.306697, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06625 s, avg_samples: 128.0, ips: 120.04724 samples/s, eta: 1:44:43
eval model:: 100%|██████████████████████████████| 18/18 [00:07<00:00,  2.50it/s]
[2022/08/25 00:25:47] ppocr INFO: cur metric, acc: 0.9759930118900347, norm_edit_dis: 0.9892101231842345, fps: 439.4374038102637
[2022/08/25 00:25:50] ppocr INFO: save best model is to ./output/rec/PPOCRV3_0.5/best_accuracy
[2022/08/25 00:25:50] ppocr INFO: best metric, acc: 0.9759930118900347, norm_edit_dis: 0.9892101231842345, fps: 439.4374038102637, best_epoch: 83, start_epoch: 78
[2022/08/25 00:26:11] ppocr INFO: epoch: [83/100], global_step: 2020, lr: 0.000121, acc: 0.945312, norm_edit_dis: 0.987254, loss: 0.379160, avg_reader_cost: 0.00175 s, avg_batch_cost: 1.06714 s, avg_samples: 128.0, ips: 119.94691 samples/s, eta: 1:44:21
[2022/08/25 00:26:33] ppocr INFO: epoch: [83/100], global_step: 2040, lr: 0.000120, acc: 0.937500, norm_edit_dis: 0.985537, loss: 0.454193, avg_reader_cost: 0.00073 s, avg_batch_cost: 1.06584 s, avg_samples: 128.0, ips: 120.09264 samples/s, eta: 1:43:58
[2022/08/25 00:26:36] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:26:59] ppocr INFO: epoch: [84/100], global_step: 2060, lr: 0.000120, acc: 0.937500, norm_edit_dis: 0.986241, loss: 0.424506, avg_reader_cost: 0.22782 s, avg_batch_cost: 1.29291 s, avg_samples: 128.0, ips: 99.00180 samples/s, eta: 1:43:49
[2022/08/25 00:27:20] ppocr INFO: epoch: [84/100], global_step: 2080, lr: 0.000119, acc: 0.937500, norm_edit_dis: 0.987190, loss: 0.462364, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.05999 s, avg_samples: 128.0, ips: 120.75609 samples/s, eta: 1:43:26
[2022/08/25 00:27:41] ppocr INFO: epoch: [84/100], global_step: 2100, lr: 0.000119, acc: 0.949219, norm_edit_dis: 0.989481, loss: 0.405806, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06053 s, avg_samples: 128.0, ips: 120.69442 samples/s, eta: 1:43:03
[2022/08/25 00:28:02] ppocr INFO: epoch: [84/100], global_step: 2120, lr: 0.000118, acc: 0.941406, norm_edit_dis: 0.987444, loss: 0.550630, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06294 s, avg_samples: 128.0, ips: 120.42039 samples/s, eta: 1:42:41
[2022/08/25 00:28:24] ppocr INFO: epoch: [84/100], global_step: 2140, lr: 0.000117, acc: 0.941406, norm_edit_dis: 0.987049, loss: 0.404486, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06467 s, avg_samples: 128.0, ips: 120.22556 samples/s, eta: 1:42:18
[2022/08/25 00:28:45] ppocr INFO: epoch: [84/100], global_step: 2160, lr: 0.000117, acc: 0.941406, norm_edit_dis: 0.988451, loss: 0.435811, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06455 s, avg_samples: 128.0, ips: 120.23906 samples/s, eta: 1:41:56
[2022/08/25 00:29:06] ppocr INFO: epoch: [84/100], global_step: 2180, lr: 0.000116, acc: 0.945312, norm_edit_dis: 0.988652, loss: 0.421111, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06449 s, avg_samples: 128.0, ips: 120.24501 samples/s, eta: 1:41:33
[2022/08/25 00:29:28] ppocr INFO: epoch: [84/100], global_step: 2200, lr: 0.000116, acc: 0.953125, norm_edit_dis: 0.987944, loss: 0.277291, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06572 s, avg_samples: 128.0, ips: 120.10647 samples/s, eta: 1:41:11
[2022/08/25 00:29:49] ppocr INFO: epoch: [84/100], global_step: 2220, lr: 0.000115, acc: 0.945312, norm_edit_dis: 0.989527, loss: 0.308942, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06731 s, avg_samples: 128.0, ips: 119.92810 samples/s, eta: 1:40:49
[2022/08/25 00:30:10] ppocr INFO: epoch: [84/100], global_step: 2240, lr: 0.000114, acc: 0.949219, norm_edit_dis: 0.990149, loss: 0.362423, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06692 s, avg_samples: 128.0, ips: 119.97117 samples/s, eta: 1:40:26
[2022/08/25 00:30:32] ppocr INFO: epoch: [84/100], global_step: 2260, lr: 0.000114, acc: 0.937500, norm_edit_dis: 0.985475, loss: 0.505254, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06548 s, avg_samples: 128.0, ips: 120.13394 samples/s, eta: 1:40:04
[2022/08/25 00:30:53] ppocr INFO: epoch: [84/100], global_step: 2280, lr: 0.000113, acc: 0.945312, norm_edit_dis: 0.987591, loss: 0.414968, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06621 s, avg_samples: 128.0, ips: 120.05194 samples/s, eta: 1:39:42
[2022/08/25 00:31:14] ppocr INFO: epoch: [84/100], global_step: 2300, lr: 0.000113, acc: 0.949219, norm_edit_dis: 0.988459, loss: 0.350465, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06607 s, avg_samples: 128.0, ips: 120.06722 samples/s, eta: 1:39:19
[2022/08/25 00:31:36] ppocr INFO: epoch: [84/100], global_step: 2320, lr: 0.000112, acc: 0.953125, norm_edit_dis: 0.988124, loss: 0.404971, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06451 s, avg_samples: 128.0, ips: 120.24298 samples/s, eta: 1:38:57
[2022/08/25 00:31:57] ppocr INFO: epoch: [84/100], global_step: 2340, lr: 0.000112, acc: 0.945312, norm_edit_dis: 0.987864, loss: 0.381122, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06379 s, avg_samples: 128.0, ips: 120.32397 samples/s, eta: 1:38:35
[2022/08/25 00:32:18] ppocr INFO: epoch: [84/100], global_step: 2360, lr: 0.000111, acc: 0.953125, norm_edit_dis: 0.988835, loss: 0.411924, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06484 s, avg_samples: 128.0, ips: 120.20629 samples/s, eta: 1:38:13
[2022/08/25 00:32:39] ppocr INFO: epoch: [84/100], global_step: 2380, lr: 0.000110, acc: 0.937500, norm_edit_dis: 0.987850, loss: 0.465640, avg_reader_cost: 0.00047 s, avg_batch_cost: 1.06562 s, avg_samples: 128.0, ips: 120.11799 samples/s, eta: 1:37:50
[2022/08/25 00:32:42] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:33:05] ppocr INFO: epoch: [85/100], global_step: 2400, lr: 0.000110, acc: 0.937500, norm_edit_dis: 0.985459, loss: 0.385063, avg_reader_cost: 0.22945 s, avg_batch_cost: 1.29807 s, avg_samples: 128.0, ips: 98.60777 samples/s, eta: 1:37:39
[2022/08/25 00:33:27] ppocr INFO: epoch: [85/100], global_step: 2420, lr: 0.000109, acc: 0.945312, norm_edit_dis: 0.988506, loss: 0.294317, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06493 s, avg_samples: 128.0, ips: 120.19574 samples/s, eta: 1:37:16
[2022/08/25 00:33:48] ppocr INFO: epoch: [85/100], global_step: 2440, lr: 0.000109, acc: 0.945312, norm_edit_dis: 0.987478, loss: 0.379200, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06265 s, avg_samples: 128.0, ips: 120.45303 samples/s, eta: 1:36:54
[2022/08/25 00:34:09] ppocr INFO: epoch: [85/100], global_step: 2460, lr: 0.000108, acc: 0.949219, norm_edit_dis: 0.986747, loss: 0.347658, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06436 s, avg_samples: 128.0, ips: 120.26055 samples/s, eta: 1:36:32
[2022/08/25 00:34:31] ppocr INFO: epoch: [85/100], global_step: 2480, lr: 0.000107, acc: 0.937500, norm_edit_dis: 0.985390, loss: 0.408098, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06472 s, avg_samples: 128.0, ips: 120.21930 samples/s, eta: 1:36:09
[2022/08/25 00:34:52] ppocr INFO: epoch: [85/100], global_step: 2500, lr: 0.000107, acc: 0.949219, norm_edit_dis: 0.990436, loss: 0.421390, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06658 s, avg_samples: 128.0, ips: 120.00961 samples/s, eta: 1:35:47
[2022/08/25 00:35:13] ppocr INFO: epoch: [85/100], global_step: 2520, lr: 0.000106, acc: 0.945312, norm_edit_dis: 0.989322, loss: 0.473354, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06754 s, avg_samples: 128.0, ips: 119.90136 samples/s, eta: 1:35:25
[2022/08/25 00:35:35] ppocr INFO: epoch: [85/100], global_step: 2540, lr: 0.000106, acc: 0.945312, norm_edit_dis: 0.989203, loss: 0.432827, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06676 s, avg_samples: 128.0, ips: 119.98909 samples/s, eta: 1:35:03
[2022/08/25 00:35:56] ppocr INFO: epoch: [85/100], global_step: 2560, lr: 0.000105, acc: 0.953125, norm_edit_dis: 0.991634, loss: 0.339455, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06658 s, avg_samples: 128.0, ips: 120.00939 samples/s, eta: 1:34:41
[2022/08/25 00:36:17] ppocr INFO: epoch: [85/100], global_step: 2580, lr: 0.000105, acc: 0.945312, norm_edit_dis: 0.986996, loss: 0.337990, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06770 s, avg_samples: 128.0, ips: 119.88364 samples/s, eta: 1:34:18
[2022/08/25 00:36:39] ppocr INFO: epoch: [85/100], global_step: 2600, lr: 0.000104, acc: 0.953125, norm_edit_dis: 0.989393, loss: 0.368683, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06718 s, avg_samples: 128.0, ips: 119.94173 samples/s, eta: 1:33:56
[2022/08/25 00:37:00] ppocr INFO: epoch: [85/100], global_step: 2620, lr: 0.000104, acc: 0.953125, norm_edit_dis: 0.988469, loss: 0.341464, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06433 s, avg_samples: 128.0, ips: 120.26375 samples/s, eta: 1:33:34
[2022/08/25 00:37:21] ppocr INFO: epoch: [85/100], global_step: 2640, lr: 0.000103, acc: 0.937500, norm_edit_dis: 0.988260, loss: 0.357465, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06370 s, avg_samples: 128.0, ips: 120.33499 samples/s, eta: 1:33:12
[2022/08/25 00:37:43] ppocr INFO: epoch: [85/100], global_step: 2660, lr: 0.000102, acc: 0.945312, norm_edit_dis: 0.989044, loss: 0.371674, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06323 s, avg_samples: 128.0, ips: 120.38770 samples/s, eta: 1:32:50
[2022/08/25 00:38:04] ppocr INFO: epoch: [85/100], global_step: 2680, lr: 0.000102, acc: 0.953125, norm_edit_dis: 0.988537, loss: 0.380005, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06478 s, avg_samples: 128.0, ips: 120.21301 samples/s, eta: 1:32:28
[2022/08/25 00:38:25] ppocr INFO: epoch: [85/100], global_step: 2700, lr: 0.000101, acc: 0.937500, norm_edit_dis: 0.987952, loss: 0.453057, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06587 s, avg_samples: 128.0, ips: 120.08929 samples/s, eta: 1:32:05
[2022/08/25 00:38:47] ppocr INFO: epoch: [85/100], global_step: 2720, lr: 0.000101, acc: 0.945312, norm_edit_dis: 0.988384, loss: 0.372552, avg_reader_cost: 0.00048 s, avg_batch_cost: 1.06679 s, avg_samples: 128.0, ips: 119.98591 samples/s, eta: 1:31:43
[2022/08/25 00:38:49] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:39:12] ppocr INFO: epoch: [86/100], global_step: 2740, lr: 0.000100, acc: 0.945312, norm_edit_dis: 0.987756, loss: 0.397138, avg_reader_cost: 0.22023 s, avg_batch_cost: 1.28630 s, avg_samples: 128.0, ips: 99.51004 samples/s, eta: 1:31:29
[2022/08/25 00:39:34] ppocr INFO: epoch: [86/100], global_step: 2760, lr: 0.000100, acc: 0.953125, norm_edit_dis: 0.990346, loss: 0.337279, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06417 s, avg_samples: 128.0, ips: 120.28112 samples/s, eta: 1:31:07
[2022/08/25 00:39:55] ppocr INFO: epoch: [86/100], global_step: 2780, lr: 0.000099, acc: 0.953125, norm_edit_dis: 0.990375, loss: 0.419321, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06597 s, avg_samples: 128.0, ips: 120.07885 samples/s, eta: 1:30:45
[2022/08/25 00:40:16] ppocr INFO: epoch: [86/100], global_step: 2800, lr: 0.000099, acc: 0.949219, norm_edit_dis: 0.990093, loss: 0.359350, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06520 s, avg_samples: 128.0, ips: 120.16471 samples/s, eta: 1:30:23
[2022/08/25 00:40:38] ppocr INFO: epoch: [86/100], global_step: 2820, lr: 0.000098, acc: 0.953125, norm_edit_dis: 0.987946, loss: 0.343449, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06629 s, avg_samples: 128.0, ips: 120.04206 samples/s, eta: 1:30:01
[2022/08/25 00:40:59] ppocr INFO: epoch: [86/100], global_step: 2840, lr: 0.000097, acc: 0.953125, norm_edit_dis: 0.991540, loss: 0.251567, avg_reader_cost: 0.00055 s, avg_batch_cost: 1.06749 s, avg_samples: 128.0, ips: 119.90731 samples/s, eta: 1:29:39
[2022/08/25 00:41:20] ppocr INFO: epoch: [86/100], global_step: 2860, lr: 0.000097, acc: 0.945312, norm_edit_dis: 0.990281, loss: 0.351908, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06784 s, avg_samples: 128.0, ips: 119.86803 samples/s, eta: 1:29:17
[2022/08/25 00:41:42] ppocr INFO: epoch: [86/100], global_step: 2880, lr: 0.000096, acc: 0.949219, norm_edit_dis: 0.989828, loss: 0.318042, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06869 s, avg_samples: 128.0, ips: 119.77319 samples/s, eta: 1:28:55
[2022/08/25 00:42:03] ppocr INFO: epoch: [86/100], global_step: 2900, lr: 0.000096, acc: 0.960937, norm_edit_dis: 0.991021, loss: 0.322881, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06969 s, avg_samples: 128.0, ips: 119.66116 samples/s, eta: 1:28:33
[2022/08/25 00:42:24] ppocr INFO: epoch: [86/100], global_step: 2920, lr: 0.000095, acc: 0.945312, norm_edit_dis: 0.987161, loss: 0.393311, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06571 s, avg_samples: 128.0, ips: 120.10738 samples/s, eta: 1:28:11
[2022/08/25 00:42:46] ppocr INFO: epoch: [86/100], global_step: 2940, lr: 0.000095, acc: 0.957031, norm_edit_dis: 0.988117, loss: 0.291313, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06673 s, avg_samples: 128.0, ips: 119.99308 samples/s, eta: 1:27:49
[2022/08/25 00:43:07] ppocr INFO: epoch: [86/100], global_step: 2960, lr: 0.000094, acc: 0.953125, norm_edit_dis: 0.991649, loss: 0.259069, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06559 s, avg_samples: 128.0, ips: 120.12132 samples/s, eta: 1:27:27
[2022/08/25 00:43:28] ppocr INFO: epoch: [86/100], global_step: 2980, lr: 0.000094, acc: 0.953125, norm_edit_dis: 0.989505, loss: 0.327935, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06644 s, avg_samples: 128.0, ips: 120.02497 samples/s, eta: 1:27:05
[2022/08/25 00:43:50] ppocr INFO: epoch: [86/100], global_step: 3000, lr: 0.000093, acc: 0.953125, norm_edit_dis: 0.988085, loss: 0.413814, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06719 s, avg_samples: 128.0, ips: 119.94160 samples/s, eta: 1:26:43
[2022/08/25 00:44:11] ppocr INFO: epoch: [86/100], global_step: 3020, lr: 0.000093, acc: 0.945312, norm_edit_dis: 0.988179, loss: 0.358817, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06659 s, avg_samples: 128.0, ips: 120.00918 samples/s, eta: 1:26:21
[2022/08/25 00:44:32] ppocr INFO: epoch: [86/100], global_step: 3040, lr: 0.000092, acc: 0.945312, norm_edit_dis: 0.988552, loss: 0.414998, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06503 s, avg_samples: 128.0, ips: 120.18424 samples/s, eta: 1:25:59
[2022/08/25 00:44:54] ppocr INFO: epoch: [86/100], global_step: 3060, lr: 0.000091, acc: 0.953125, norm_edit_dis: 0.991450, loss: 0.289862, avg_reader_cost: 0.00057 s, avg_batch_cost: 1.06476 s, avg_samples: 128.0, ips: 120.21448 samples/s, eta: 1:25:37
[2022/08/25 00:44:57] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:45:20] ppocr INFO: epoch: [87/100], global_step: 3080, lr: 0.000091, acc: 0.941406, norm_edit_dis: 0.987796, loss: 0.491427, avg_reader_cost: 0.22936 s, avg_batch_cost: 1.29610 s, avg_samples: 128.0, ips: 98.75766 samples/s, eta: 1:25:22
[2022/08/25 00:45:41] ppocr INFO: epoch: [87/100], global_step: 3100, lr: 0.000090, acc: 0.953125, norm_edit_dis: 0.990194, loss: 0.308192, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06090 s, avg_samples: 128.0, ips: 120.65192 samples/s, eta: 1:24:59
[2022/08/25 00:46:02] ppocr INFO: epoch: [87/100], global_step: 3120, lr: 0.000090, acc: 0.949219, norm_edit_dis: 0.991161, loss: 0.371672, avg_reader_cost: 0.00051 s, avg_batch_cost: 1.06137 s, avg_samples: 128.0, ips: 120.59886 samples/s, eta: 1:24:37
[2022/08/25 00:46:23] ppocr INFO: epoch: [87/100], global_step: 3140, lr: 0.000089, acc: 0.953125, norm_edit_dis: 0.988681, loss: 0.365214, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06302 s, avg_samples: 128.0, ips: 120.41210 samples/s, eta: 1:24:15
[2022/08/25 00:46:45] ppocr INFO: epoch: [87/100], global_step: 3160, lr: 0.000089, acc: 0.953125, norm_edit_dis: 0.989508, loss: 0.325195, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06333 s, avg_samples: 128.0, ips: 120.37599 samples/s, eta: 1:23:53
[2022/08/25 00:47:06] ppocr INFO: epoch: [87/100], global_step: 3180, lr: 0.000088, acc: 0.945312, norm_edit_dis: 0.989409, loss: 0.418040, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06222 s, avg_samples: 128.0, ips: 120.50235 samples/s, eta: 1:23:31
[2022/08/25 00:47:27] ppocr INFO: epoch: [87/100], global_step: 3200, lr: 0.000088, acc: 0.945312, norm_edit_dis: 0.988281, loss: 0.364804, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06322 s, avg_samples: 128.0, ips: 120.38956 samples/s, eta: 1:23:09
[2022/08/25 00:47:48] ppocr INFO: epoch: [87/100], global_step: 3220, lr: 0.000087, acc: 0.945312, norm_edit_dis: 0.989723, loss: 0.323088, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06419 s, avg_samples: 128.0, ips: 120.27880 samples/s, eta: 1:22:47
[2022/08/25 00:48:10] ppocr INFO: epoch: [87/100], global_step: 3240, lr: 0.000087, acc: 0.960937, norm_edit_dis: 0.989110, loss: 0.312414, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06360 s, avg_samples: 128.0, ips: 120.34545 samples/s, eta: 1:22:25
[2022/08/25 00:48:31] ppocr INFO: epoch: [87/100], global_step: 3260, lr: 0.000086, acc: 0.960937, norm_edit_dis: 0.993464, loss: 0.325524, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06474 s, avg_samples: 128.0, ips: 120.21742 samples/s, eta: 1:22:03
[2022/08/25 00:48:52] ppocr INFO: epoch: [87/100], global_step: 3280, lr: 0.000086, acc: 0.953125, norm_edit_dis: 0.990122, loss: 0.345493, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06334 s, avg_samples: 128.0, ips: 120.37539 samples/s, eta: 1:21:41
[2022/08/25 00:49:14] ppocr INFO: epoch: [87/100], global_step: 3300, lr: 0.000085, acc: 0.953125, norm_edit_dis: 0.990783, loss: 0.295240, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06273 s, avg_samples: 128.0, ips: 120.44458 samples/s, eta: 1:21:19
[2022/08/25 00:49:35] ppocr INFO: epoch: [87/100], global_step: 3320, lr: 0.000085, acc: 0.960937, norm_edit_dis: 0.992832, loss: 0.280843, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06380 s, avg_samples: 128.0, ips: 120.32316 samples/s, eta: 1:20:57
[2022/08/25 00:49:56] ppocr INFO: epoch: [87/100], global_step: 3340, lr: 0.000084, acc: 0.945312, norm_edit_dis: 0.989641, loss: 0.317477, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06385 s, avg_samples: 128.0, ips: 120.31780 samples/s, eta: 1:20:35
[2022/08/25 00:50:17] ppocr INFO: epoch: [87/100], global_step: 3360, lr: 0.000084, acc: 0.957031, norm_edit_dis: 0.992024, loss: 0.294300, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06449 s, avg_samples: 128.0, ips: 120.24484 samples/s, eta: 1:20:13
[2022/08/25 00:50:39] ppocr INFO: epoch: [87/100], global_step: 3380, lr: 0.000083, acc: 0.960937, norm_edit_dis: 0.992992, loss: 0.250903, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06509 s, avg_samples: 128.0, ips: 120.17751 samples/s, eta: 1:19:51
[2022/08/25 00:51:00] ppocr INFO: epoch: [87/100], global_step: 3400, lr: 0.000083, acc: 0.949219, norm_edit_dis: 0.990176, loss: 0.333674, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06425 s, avg_samples: 128.0, ips: 120.27289 samples/s, eta: 1:19:29
[2022/08/25 00:51:03] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:51:26] ppocr INFO: epoch: [88/100], global_step: 3420, lr: 0.000082, acc: 0.960937, norm_edit_dis: 0.991666, loss: 0.344220, avg_reader_cost: 0.22826 s, avg_batch_cost: 1.29499 s, avg_samples: 128.0, ips: 98.84236 samples/s, eta: 1:19:13
[2022/08/25 00:51:47] ppocr INFO: epoch: [88/100], global_step: 3440, lr: 0.000082, acc: 0.949219, norm_edit_dis: 0.989441, loss: 0.313812, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06329 s, avg_samples: 128.0, ips: 120.38149 samples/s, eta: 1:18:51
[2022/08/25 00:52:08] ppocr INFO: epoch: [88/100], global_step: 3460, lr: 0.000081, acc: 0.949219, norm_edit_dis: 0.990511, loss: 0.292587, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06227 s, avg_samples: 128.0, ips: 120.49670 samples/s, eta: 1:18:29
[2022/08/25 00:52:30] ppocr INFO: epoch: [88/100], global_step: 3480, lr: 0.000081, acc: 0.953125, norm_edit_dis: 0.989968, loss: 0.276255, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06295 s, avg_samples: 128.0, ips: 120.42003 samples/s, eta: 1:18:07
[2022/08/25 00:52:51] ppocr INFO: epoch: [88/100], global_step: 3500, lr: 0.000080, acc: 0.945312, norm_edit_dis: 0.987344, loss: 0.346320, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06294 s, avg_samples: 128.0, ips: 120.42038 samples/s, eta: 1:17:45
[2022/08/25 00:53:12] ppocr INFO: epoch: [88/100], global_step: 3520, lr: 0.000080, acc: 0.945312, norm_edit_dis: 0.988240, loss: 0.372822, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06406 s, avg_samples: 128.0, ips: 120.29383 samples/s, eta: 1:17:23
[2022/08/25 00:53:34] ppocr INFO: epoch: [88/100], global_step: 3540, lr: 0.000079, acc: 0.960937, norm_edit_dis: 0.990348, loss: 0.287657, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06332 s, avg_samples: 128.0, ips: 120.37739 samples/s, eta: 1:17:01
[2022/08/25 00:53:55] ppocr INFO: epoch: [88/100], global_step: 3560, lr: 0.000079, acc: 0.960937, norm_edit_dis: 0.992180, loss: 0.345466, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06346 s, avg_samples: 128.0, ips: 120.36199 samples/s, eta: 1:16:39
[2022/08/25 00:54:16] ppocr INFO: epoch: [88/100], global_step: 3580, lr: 0.000078, acc: 0.953125, norm_edit_dis: 0.990087, loss: 0.405446, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06424 s, avg_samples: 128.0, ips: 120.27394 samples/s, eta: 1:16:17
[2022/08/25 00:54:37] ppocr INFO: epoch: [88/100], global_step: 3600, lr: 0.000078, acc: 0.953125, norm_edit_dis: 0.991459, loss: 0.261606, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06506 s, avg_samples: 128.0, ips: 120.18054 samples/s, eta: 1:15:55
[2022/08/25 00:54:59] ppocr INFO: epoch: [88/100], global_step: 3620, lr: 0.000077, acc: 0.945312, norm_edit_dis: 0.987183, loss: 0.351831, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06475 s, avg_samples: 128.0, ips: 120.21625 samples/s, eta: 1:15:33
[2022/08/25 00:55:20] ppocr INFO: epoch: [88/100], global_step: 3640, lr: 0.000077, acc: 0.953125, norm_edit_dis: 0.990328, loss: 0.294807, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06372 s, avg_samples: 128.0, ips: 120.33232 samples/s, eta: 1:15:11
[2022/08/25 00:55:41] ppocr INFO: epoch: [88/100], global_step: 3660, lr: 0.000076, acc: 0.949219, norm_edit_dis: 0.990580, loss: 0.359036, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06157 s, avg_samples: 128.0, ips: 120.57626 samples/s, eta: 1:14:49
[2022/08/25 00:56:03] ppocr INFO: epoch: [88/100], global_step: 3680, lr: 0.000076, acc: 0.953125, norm_edit_dis: 0.989247, loss: 0.374247, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06082 s, avg_samples: 128.0, ips: 120.66115 samples/s, eta: 1:14:27
[2022/08/25 00:56:24] ppocr INFO: epoch: [88/100], global_step: 3700, lr: 0.000075, acc: 0.945312, norm_edit_dis: 0.989501, loss: 0.397247, avg_reader_cost: 0.00040 s, avg_batch_cost: 1.06045 s, avg_samples: 128.0, ips: 120.70399 samples/s, eta: 1:14:05
[2022/08/25 00:56:45] ppocr INFO: epoch: [88/100], global_step: 3720, lr: 0.000075, acc: 0.957031, norm_edit_dis: 0.990916, loss: 0.316079, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06127 s, avg_samples: 128.0, ips: 120.61029 samples/s, eta: 1:13:43
[2022/08/25 00:57:06] ppocr INFO: epoch: [88/100], global_step: 3740, lr: 0.000074, acc: 0.957031, norm_edit_dis: 0.990803, loss: 0.286964, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06202 s, avg_samples: 128.0, ips: 120.52492 samples/s, eta: 1:13:21
[2022/08/25 00:57:09] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 00:57:32] ppocr INFO: epoch: [89/100], global_step: 3760, lr: 0.000074, acc: 0.953125, norm_edit_dis: 0.990273, loss: 0.280679, avg_reader_cost: 0.21749 s, avg_batch_cost: 1.28395 s, avg_samples: 128.0, ips: 99.69206 samples/s, eta: 1:13:04
[2022/08/25 00:57:53] ppocr INFO: epoch: [89/100], global_step: 3780, lr: 0.000073, acc: 0.957031, norm_edit_dis: 0.992356, loss: 0.287999, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06294 s, avg_samples: 128.0, ips: 120.42097 samples/s, eta: 1:12:42
[2022/08/25 00:58:14] ppocr INFO: epoch: [89/100], global_step: 3800, lr: 0.000073, acc: 0.953125, norm_edit_dis: 0.990955, loss: 0.320984, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06403 s, avg_samples: 128.0, ips: 120.29752 samples/s, eta: 1:12:20
[2022/08/25 00:58:36] ppocr INFO: epoch: [89/100], global_step: 3820, lr: 0.000072, acc: 0.949219, norm_edit_dis: 0.990891, loss: 0.285772, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06350 s, avg_samples: 128.0, ips: 120.35677 samples/s, eta: 1:11:58
[2022/08/25 00:58:57] ppocr INFO: epoch: [89/100], global_step: 3840, lr: 0.000072, acc: 0.953125, norm_edit_dis: 0.990005, loss: 0.334990, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06446 s, avg_samples: 128.0, ips: 120.24920 samples/s, eta: 1:11:36
[2022/08/25 00:59:18] ppocr INFO: epoch: [89/100], global_step: 3860, lr: 0.000071, acc: 0.957031, norm_edit_dis: 0.990218, loss: 0.447327, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06476 s, avg_samples: 128.0, ips: 120.21536 samples/s, eta: 1:11:14
[2022/08/25 00:59:40] ppocr INFO: epoch: [89/100], global_step: 3880, lr: 0.000071, acc: 0.960937, norm_edit_dis: 0.989446, loss: 0.331613, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06382 s, avg_samples: 128.0, ips: 120.32123 samples/s, eta: 1:10:52
[2022/08/25 01:00:01] ppocr INFO: epoch: [89/100], global_step: 3900, lr: 0.000070, acc: 0.960937, norm_edit_dis: 0.993350, loss: 0.291088, avg_reader_cost: 0.00051 s, avg_batch_cost: 1.06635 s, avg_samples: 128.0, ips: 120.03594 samples/s, eta: 1:10:30
[2022/08/25 01:00:22] ppocr INFO: epoch: [89/100], global_step: 3920, lr: 0.000070, acc: 0.941406, norm_edit_dis: 0.986082, loss: 0.355709, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06419 s, avg_samples: 128.0, ips: 120.27902 samples/s, eta: 1:10:09
[2022/08/25 01:00:44] ppocr INFO: epoch: [89/100], global_step: 3940, lr: 0.000069, acc: 0.949219, norm_edit_dis: 0.990910, loss: 0.332387, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06403 s, avg_samples: 128.0, ips: 120.29723 samples/s, eta: 1:09:47
[2022/08/25 01:01:05] ppocr INFO: epoch: [89/100], global_step: 3960, lr: 0.000069, acc: 0.953125, norm_edit_dis: 0.991890, loss: 0.308850, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06430 s, avg_samples: 128.0, ips: 120.26649 samples/s, eta: 1:09:25
[2022/08/25 01:01:26] ppocr INFO: epoch: [89/100], global_step: 3980, lr: 0.000068, acc: 0.960937, norm_edit_dis: 0.992621, loss: 0.264161, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06472 s, avg_samples: 128.0, ips: 120.21906 samples/s, eta: 1:09:03
[2022/08/25 01:01:47] ppocr INFO: epoch: [89/100], global_step: 4000, lr: 0.000068, acc: 0.953125, norm_edit_dis: 0.989346, loss: 0.279698, avg_reader_cost: 0.00061 s, avg_batch_cost: 1.06409 s, avg_samples: 128.0, ips: 120.29034 samples/s, eta: 1:08:41
eval model:: 100%|██████████████████████████████| 18/18 [00:07<00:00,  2.52it/s]
[2022/08/25 01:01:55] ppocr INFO: cur metric, acc: 0.9746835400493952, norm_edit_dis: 0.9886849203313461, fps: 439.69697596722324
[2022/08/25 01:01:55] ppocr INFO: best metric, acc: 0.9759930118900347, norm_edit_dis: 0.9892101231842345, fps: 439.4374038102637, best_epoch: 83, start_epoch: 78
[2022/08/25 01:02:17] ppocr INFO: epoch: [89/100], global_step: 4020, lr: 0.000068, acc: 0.953125, norm_edit_dis: 0.991968, loss: 0.320792, avg_reader_cost: 0.00164 s, avg_batch_cost: 1.06175 s, avg_samples: 128.0, ips: 120.55522 samples/s, eta: 1:08:19
[2022/08/25 01:02:38] ppocr INFO: epoch: [89/100], global_step: 4040, lr: 0.000067, acc: 0.953125, norm_edit_dis: 0.990107, loss: 0.385939, avg_reader_cost: 0.00050 s, avg_batch_cost: 1.06291 s, avg_samples: 128.0, ips: 120.42393 samples/s, eta: 1:07:57
[2022/08/25 01:02:59] ppocr INFO: epoch: [89/100], global_step: 4060, lr: 0.000067, acc: 0.949219, norm_edit_dis: 0.989447, loss: 0.415306, avg_reader_cost: 0.00040 s, avg_batch_cost: 1.06389 s, avg_samples: 128.0, ips: 120.31369 samples/s, eta: 1:07:35
[2022/08/25 01:03:20] ppocr INFO: epoch: [89/100], global_step: 4080, lr: 0.000066, acc: 0.945312, norm_edit_dis: 0.989529, loss: 0.284462, avg_reader_cost: 0.00044 s, avg_batch_cost: 1.06427 s, avg_samples: 128.0, ips: 120.27003 samples/s, eta: 1:07:14
[2022/08/25 01:03:23] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:03:46] ppocr INFO: epoch: [90/100], global_step: 4100, lr: 0.000066, acc: 0.957031, norm_edit_dis: 0.991265, loss: 0.285601, avg_reader_cost: 0.22447 s, avg_batch_cost: 1.29028 s, avg_samples: 128.0, ips: 99.20292 samples/s, eta: 1:06:56
[2022/08/25 01:04:07] ppocr INFO: epoch: [90/100], global_step: 4120, lr: 0.000065, acc: 0.957031, norm_edit_dis: 0.990650, loss: 0.258992, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06249 s, avg_samples: 128.0, ips: 120.47149 samples/s, eta: 1:06:34
[2022/08/25 01:04:29] ppocr INFO: epoch: [90/100], global_step: 4140, lr: 0.000065, acc: 0.960937, norm_edit_dis: 0.990938, loss: 0.353663, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06178 s, avg_samples: 128.0, ips: 120.55269 samples/s, eta: 1:06:12
[2022/08/25 01:04:50] ppocr INFO: epoch: [90/100], global_step: 4160, lr: 0.000064, acc: 0.960937, norm_edit_dis: 0.992280, loss: 0.241399, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06155 s, avg_samples: 128.0, ips: 120.57830 samples/s, eta: 1:05:50
[2022/08/25 01:05:11] ppocr INFO: epoch: [90/100], global_step: 4180, lr: 0.000064, acc: 0.960937, norm_edit_dis: 0.989812, loss: 0.273452, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06174 s, avg_samples: 128.0, ips: 120.55723 samples/s, eta: 1:05:28
[2022/08/25 01:05:32] ppocr INFO: epoch: [90/100], global_step: 4200, lr: 0.000063, acc: 0.953125, norm_edit_dis: 0.988937, loss: 0.267283, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06115 s, avg_samples: 128.0, ips: 120.62437 samples/s, eta: 1:05:06
[2022/08/25 01:05:54] ppocr INFO: epoch: [90/100], global_step: 4220, lr: 0.000063, acc: 0.953125, norm_edit_dis: 0.991175, loss: 0.289410, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06226 s, avg_samples: 128.0, ips: 120.49829 samples/s, eta: 1:04:44
[2022/08/25 01:06:15] ppocr INFO: epoch: [90/100], global_step: 4240, lr: 0.000063, acc: 0.945312, norm_edit_dis: 0.990797, loss: 0.259348, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06087 s, avg_samples: 128.0, ips: 120.65561 samples/s, eta: 1:04:23
[2022/08/25 01:06:36] ppocr INFO: epoch: [90/100], global_step: 4260, lr: 0.000062, acc: 0.957031, norm_edit_dis: 0.992608, loss: 0.335772, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06305 s, avg_samples: 128.0, ips: 120.40774 samples/s, eta: 1:04:01
[2022/08/25 01:06:57] ppocr INFO: epoch: [90/100], global_step: 4280, lr: 0.000062, acc: 0.949219, norm_edit_dis: 0.989318, loss: 0.393096, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06346 s, avg_samples: 128.0, ips: 120.36229 samples/s, eta: 1:03:39
[2022/08/25 01:07:19] ppocr INFO: epoch: [90/100], global_step: 4300, lr: 0.000061, acc: 0.941406, norm_edit_dis: 0.987751, loss: 0.387869, avg_reader_cost: 0.00040 s, avg_batch_cost: 1.06317 s, avg_samples: 128.0, ips: 120.39442 samples/s, eta: 1:03:17
[2022/08/25 01:07:40] ppocr INFO: epoch: [90/100], global_step: 4320, lr: 0.000061, acc: 0.953125, norm_edit_dis: 0.991096, loss: 0.280916, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06354 s, avg_samples: 128.0, ips: 120.35226 samples/s, eta: 1:02:55
[2022/08/25 01:08:01] ppocr INFO: epoch: [90/100], global_step: 4340, lr: 0.000060, acc: 0.957031, norm_edit_dis: 0.990515, loss: 0.324798, avg_reader_cost: 0.00051 s, avg_batch_cost: 1.06375 s, avg_samples: 128.0, ips: 120.32864 samples/s, eta: 1:02:33
[2022/08/25 01:08:23] ppocr INFO: epoch: [90/100], global_step: 4360, lr: 0.000060, acc: 0.953125, norm_edit_dis: 0.991758, loss: 0.329539, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06153 s, avg_samples: 128.0, ips: 120.58057 samples/s, eta: 1:02:12
[2022/08/25 01:08:44] ppocr INFO: epoch: [90/100], global_step: 4380, lr: 0.000059, acc: 0.945312, norm_edit_dis: 0.989810, loss: 0.282879, avg_reader_cost: 0.00040 s, avg_batch_cost: 1.06095 s, avg_samples: 128.0, ips: 120.64647 samples/s, eta: 1:01:50
[2022/08/25 01:09:05] ppocr INFO: epoch: [90/100], global_step: 4400, lr: 0.000059, acc: 0.960937, norm_edit_dis: 0.991188, loss: 0.285911, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06201 s, avg_samples: 128.0, ips: 120.52585 samples/s, eta: 1:01:28
[2022/08/25 01:09:26] ppocr INFO: epoch: [90/100], global_step: 4420, lr: 0.000059, acc: 0.964844, norm_edit_dis: 0.993843, loss: 0.297063, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06278 s, avg_samples: 128.0, ips: 120.43920 samples/s, eta: 1:01:06
[2022/08/25 01:09:29] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:09:30] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/iter_epoch_90
[2022/08/25 01:09:53] ppocr INFO: epoch: [91/100], global_step: 4440, lr: 0.000058, acc: 0.953125, norm_edit_dis: 0.991410, loss: 0.265881, avg_reader_cost: 0.26539 s, avg_batch_cost: 1.33118 s, avg_samples: 128.0, ips: 96.15536 samples/s, eta: 1:00:48
[2022/08/25 01:10:14] ppocr INFO: epoch: [91/100], global_step: 4460, lr: 0.000058, acc: 0.953125, norm_edit_dis: 0.990580, loss: 0.366052, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06058 s, avg_samples: 128.0, ips: 120.68916 samples/s, eta: 1:00:26
[2022/08/25 01:10:35] ppocr INFO: epoch: [91/100], global_step: 4480, lr: 0.000057, acc: 0.960937, norm_edit_dis: 0.992895, loss: 0.312116, avg_reader_cost: 0.00051 s, avg_batch_cost: 1.06137 s, avg_samples: 128.0, ips: 120.59851 samples/s, eta: 1:00:05
[2022/08/25 01:10:57] ppocr INFO: epoch: [91/100], global_step: 4500, lr: 0.000057, acc: 0.960937, norm_edit_dis: 0.991822, loss: 0.242492, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06234 s, avg_samples: 128.0, ips: 120.48854 samples/s, eta: 0:59:43
[2022/08/25 01:11:18] ppocr INFO: epoch: [91/100], global_step: 4520, lr: 0.000056, acc: 0.945312, norm_edit_dis: 0.990241, loss: 0.324586, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06296 s, avg_samples: 128.0, ips: 120.41832 samples/s, eta: 0:59:21
[2022/08/25 01:11:39] ppocr INFO: epoch: [91/100], global_step: 4540, lr: 0.000056, acc: 0.953125, norm_edit_dis: 0.991688, loss: 0.327450, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06275 s, avg_samples: 128.0, ips: 120.44261 samples/s, eta: 0:58:59
[2022/08/25 01:12:00] ppocr INFO: epoch: [91/100], global_step: 4560, lr: 0.000056, acc: 0.960937, norm_edit_dis: 0.991513, loss: 0.268472, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06301 s, avg_samples: 128.0, ips: 120.41284 samples/s, eta: 0:58:37
[2022/08/25 01:12:22] ppocr INFO: epoch: [91/100], global_step: 4580, lr: 0.000055, acc: 0.953125, norm_edit_dis: 0.991825, loss: 0.353768, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06295 s, avg_samples: 128.0, ips: 120.41936 samples/s, eta: 0:58:15
[2022/08/25 01:12:43] ppocr INFO: epoch: [91/100], global_step: 4600, lr: 0.000055, acc: 0.968750, norm_edit_dis: 0.994102, loss: 0.231029, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06226 s, avg_samples: 128.0, ips: 120.49832 samples/s, eta: 0:57:54
[2022/08/25 01:13:04] ppocr INFO: epoch: [91/100], global_step: 4620, lr: 0.000054, acc: 0.957031, norm_edit_dis: 0.989146, loss: 0.295479, avg_reader_cost: 0.00040 s, avg_batch_cost: 1.06192 s, avg_samples: 128.0, ips: 120.53592 samples/s, eta: 0:57:32
[2022/08/25 01:13:25] ppocr INFO: epoch: [91/100], global_step: 4640, lr: 0.000054, acc: 0.953125, norm_edit_dis: 0.991354, loss: 0.306652, avg_reader_cost: 0.00050 s, avg_batch_cost: 1.06240 s, avg_samples: 128.0, ips: 120.48176 samples/s, eta: 0:57:10
[2022/08/25 01:13:47] ppocr INFO: epoch: [91/100], global_step: 4660, lr: 0.000053, acc: 0.953125, norm_edit_dis: 0.991777, loss: 0.274498, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06367 s, avg_samples: 128.0, ips: 120.33761 samples/s, eta: 0:56:48
[2022/08/25 01:14:08] ppocr INFO: epoch: [91/100], global_step: 4680, lr: 0.000053, acc: 0.957031, norm_edit_dis: 0.991449, loss: 0.329435, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06530 s, avg_samples: 128.0, ips: 120.15367 samples/s, eta: 0:56:27
[2022/08/25 01:14:29] ppocr INFO: epoch: [91/100], global_step: 4700, lr: 0.000053, acc: 0.953125, norm_edit_dis: 0.989967, loss: 0.263105, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06446 s, avg_samples: 128.0, ips: 120.24863 samples/s, eta: 0:56:05
[2022/08/25 01:14:51] ppocr INFO: epoch: [91/100], global_step: 4720, lr: 0.000052, acc: 0.949219, norm_edit_dis: 0.990470, loss: 0.416078, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06499 s, avg_samples: 128.0, ips: 120.18863 samples/s, eta: 0:55:43
[2022/08/25 01:15:12] ppocr INFO: epoch: [91/100], global_step: 4740, lr: 0.000052, acc: 0.953125, norm_edit_dis: 0.988172, loss: 0.285486, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06622 s, avg_samples: 128.0, ips: 120.04975 samples/s, eta: 0:55:21
[2022/08/25 01:15:33] ppocr INFO: epoch: [91/100], global_step: 4760, lr: 0.000051, acc: 0.964844, norm_edit_dis: 0.992307, loss: 0.262862, avg_reader_cost: 0.00046 s, avg_batch_cost: 1.06585 s, avg_samples: 128.0, ips: 120.09176 samples/s, eta: 0:55:00
[2022/08/25 01:15:36] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:15:59] ppocr INFO: epoch: [92/100], global_step: 4780, lr: 0.000051, acc: 0.953125, norm_edit_dis: 0.992616, loss: 0.289985, avg_reader_cost: 0.22062 s, avg_batch_cost: 1.28530 s, avg_samples: 128.0, ips: 99.58773 samples/s, eta: 0:54:41
[2022/08/25 01:16:20] ppocr INFO: epoch: [92/100], global_step: 4800, lr: 0.000051, acc: 0.968750, norm_edit_dis: 0.993604, loss: 0.224456, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06126 s, avg_samples: 128.0, ips: 120.61173 samples/s, eta: 0:54:19
[2022/08/25 01:16:41] ppocr INFO: epoch: [92/100], global_step: 4820, lr: 0.000050, acc: 0.953125, norm_edit_dis: 0.989131, loss: 0.341182, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06189 s, avg_samples: 128.0, ips: 120.54017 samples/s, eta: 0:53:57
[2022/08/25 01:17:03] ppocr INFO: epoch: [92/100], global_step: 4840, lr: 0.000050, acc: 0.964844, norm_edit_dis: 0.992587, loss: 0.329442, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06181 s, avg_samples: 128.0, ips: 120.54939 samples/s, eta: 0:53:35
[2022/08/25 01:17:24] ppocr INFO: epoch: [92/100], global_step: 4860, lr: 0.000049, acc: 0.960937, norm_edit_dis: 0.990167, loss: 0.305318, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06270 s, avg_samples: 128.0, ips: 120.44743 samples/s, eta: 0:53:13
[2022/08/25 01:17:45] ppocr INFO: epoch: [92/100], global_step: 4880, lr: 0.000049, acc: 0.960937, norm_edit_dis: 0.992364, loss: 0.248367, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06449 s, avg_samples: 128.0, ips: 120.24585 samples/s, eta: 0:52:52
[2022/08/25 01:18:07] ppocr INFO: epoch: [92/100], global_step: 4900, lr: 0.000049, acc: 0.953125, norm_edit_dis: 0.990363, loss: 0.362381, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06372 s, avg_samples: 128.0, ips: 120.33194 samples/s, eta: 0:52:30
[2022/08/25 01:18:28] ppocr INFO: epoch: [92/100], global_step: 4920, lr: 0.000048, acc: 0.960937, norm_edit_dis: 0.992101, loss: 0.218935, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06437 s, avg_samples: 128.0, ips: 120.25939 samples/s, eta: 0:52:08
[2022/08/25 01:18:49] ppocr INFO: epoch: [92/100], global_step: 4940, lr: 0.000048, acc: 0.957031, norm_edit_dis: 0.990491, loss: 0.283849, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06623 s, avg_samples: 128.0, ips: 120.04863 samples/s, eta: 0:51:46
[2022/08/25 01:19:11] ppocr INFO: epoch: [92/100], global_step: 4960, lr: 0.000047, acc: 0.957031, norm_edit_dis: 0.993097, loss: 0.250722, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06467 s, avg_samples: 128.0, ips: 120.22552 samples/s, eta: 0:51:25
[2022/08/25 01:19:32] ppocr INFO: epoch: [92/100], global_step: 4980, lr: 0.000047, acc: 0.957031, norm_edit_dis: 0.990603, loss: 0.272040, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06434 s, avg_samples: 128.0, ips: 120.26266 samples/s, eta: 0:51:03
[2022/08/25 01:19:53] ppocr INFO: epoch: [92/100], global_step: 5000, lr: 0.000047, acc: 0.968750, norm_edit_dis: 0.995386, loss: 0.217489, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06403 s, avg_samples: 128.0, ips: 120.29735 samples/s, eta: 0:50:41
[2022/08/25 01:20:14] ppocr INFO: epoch: [92/100], global_step: 5020, lr: 0.000046, acc: 0.957031, norm_edit_dis: 0.989966, loss: 0.296482, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06396 s, avg_samples: 128.0, ips: 120.30565 samples/s, eta: 0:50:19
[2022/08/25 01:20:36] ppocr INFO: epoch: [92/100], global_step: 5040, lr: 0.000046, acc: 0.964844, norm_edit_dis: 0.992013, loss: 0.286132, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06377 s, avg_samples: 128.0, ips: 120.32640 samples/s, eta: 0:49:58
[2022/08/25 01:20:57] ppocr INFO: epoch: [92/100], global_step: 5060, lr: 0.000045, acc: 0.953125, norm_edit_dis: 0.990239, loss: 0.379255, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06549 s, avg_samples: 128.0, ips: 120.13267 samples/s, eta: 0:49:36
[2022/08/25 01:21:18] ppocr INFO: epoch: [92/100], global_step: 5080, lr: 0.000045, acc: 0.957031, norm_edit_dis: 0.992963, loss: 0.320325, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06551 s, avg_samples: 128.0, ips: 120.13004 samples/s, eta: 0:49:14
[2022/08/25 01:21:40] ppocr INFO: epoch: [92/100], global_step: 5100, lr: 0.000045, acc: 0.957031, norm_edit_dis: 0.991559, loss: 0.276582, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06453 s, avg_samples: 128.0, ips: 120.24131 samples/s, eta: 0:48:53
[2022/08/25 01:21:42] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:22:05] ppocr INFO: epoch: [93/100], global_step: 5120, lr: 0.000044, acc: 0.957031, norm_edit_dis: 0.992191, loss: 0.317345, avg_reader_cost: 0.22442 s, avg_batch_cost: 1.28988 s, avg_samples: 128.0, ips: 99.23386 samples/s, eta: 0:48:33
[2022/08/25 01:22:27] ppocr INFO: epoch: [93/100], global_step: 5140, lr: 0.000044, acc: 0.960937, norm_edit_dis: 0.990020, loss: 0.335901, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06100 s, avg_samples: 128.0, ips: 120.64075 samples/s, eta: 0:48:11
[2022/08/25 01:22:48] ppocr INFO: epoch: [93/100], global_step: 5160, lr: 0.000044, acc: 0.960937, norm_edit_dis: 0.991981, loss: 0.323533, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06136 s, avg_samples: 128.0, ips: 120.60049 samples/s, eta: 0:47:50
[2022/08/25 01:23:09] ppocr INFO: epoch: [93/100], global_step: 5180, lr: 0.000043, acc: 0.953125, norm_edit_dis: 0.991587, loss: 0.268809, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06254 s, avg_samples: 128.0, ips: 120.46574 samples/s, eta: 0:47:28
[2022/08/25 01:23:30] ppocr INFO: epoch: [93/100], global_step: 5200, lr: 0.000043, acc: 0.960937, norm_edit_dis: 0.992746, loss: 0.274124, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06385 s, avg_samples: 128.0, ips: 120.31780 samples/s, eta: 0:47:06
[2022/08/25 01:23:52] ppocr INFO: epoch: [93/100], global_step: 5220, lr: 0.000042, acc: 0.960937, norm_edit_dis: 0.992259, loss: 0.279542, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06346 s, avg_samples: 128.0, ips: 120.36202 samples/s, eta: 0:46:45
[2022/08/25 01:24:13] ppocr INFO: epoch: [93/100], global_step: 5240, lr: 0.000042, acc: 0.960937, norm_edit_dis: 0.991527, loss: 0.253189, avg_reader_cost: 0.00057 s, avg_batch_cost: 1.06484 s, avg_samples: 128.0, ips: 120.20598 samples/s, eta: 0:46:23
[2022/08/25 01:24:34] ppocr INFO: epoch: [93/100], global_step: 5260, lr: 0.000042, acc: 0.960937, norm_edit_dis: 0.992181, loss: 0.312224, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06657 s, avg_samples: 128.0, ips: 120.01134 samples/s, eta: 0:46:01
[2022/08/25 01:24:56] ppocr INFO: epoch: [93/100], global_step: 5280, lr: 0.000041, acc: 0.968750, norm_edit_dis: 0.993326, loss: 0.289752, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06550 s, avg_samples: 128.0, ips: 120.13103 samples/s, eta: 0:45:39
[2022/08/25 01:25:17] ppocr INFO: epoch: [93/100], global_step: 5300, lr: 0.000041, acc: 0.957031, norm_edit_dis: 0.990542, loss: 0.297336, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06608 s, avg_samples: 128.0, ips: 120.06564 samples/s, eta: 0:45:18
[2022/08/25 01:25:38] ppocr INFO: epoch: [93/100], global_step: 5320, lr: 0.000041, acc: 0.960937, norm_edit_dis: 0.992066, loss: 0.324945, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06869 s, avg_samples: 128.0, ips: 119.77289 samples/s, eta: 0:44:56
[2022/08/25 01:26:00] ppocr INFO: epoch: [93/100], global_step: 5340, lr: 0.000040, acc: 0.968750, norm_edit_dis: 0.993390, loss: 0.225628, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06766 s, avg_samples: 128.0, ips: 119.88785 samples/s, eta: 0:44:34
[2022/08/25 01:26:21] ppocr INFO: epoch: [93/100], global_step: 5360, lr: 0.000040, acc: 0.960937, norm_edit_dis: 0.991748, loss: 0.251723, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06582 s, avg_samples: 128.0, ips: 120.09495 samples/s, eta: 0:44:13
[2022/08/25 01:26:42] ppocr INFO: epoch: [93/100], global_step: 5380, lr: 0.000039, acc: 0.960937, norm_edit_dis: 0.992080, loss: 0.277655, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06492 s, avg_samples: 128.0, ips: 120.19739 samples/s, eta: 0:43:51
[2022/08/25 01:27:04] ppocr INFO: epoch: [93/100], global_step: 5400, lr: 0.000039, acc: 0.960937, norm_edit_dis: 0.993525, loss: 0.270362, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06401 s, avg_samples: 128.0, ips: 120.29934 samples/s, eta: 0:43:29
[2022/08/25 01:27:25] ppocr INFO: epoch: [93/100], global_step: 5420, lr: 0.000039, acc: 0.960937, norm_edit_dis: 0.992942, loss: 0.283646, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06575 s, avg_samples: 128.0, ips: 120.10370 samples/s, eta: 0:43:08
[2022/08/25 01:27:46] ppocr INFO: epoch: [93/100], global_step: 5440, lr: 0.000038, acc: 0.957031, norm_edit_dis: 0.991087, loss: 0.284859, avg_reader_cost: 0.00046 s, avg_batch_cost: 1.06515 s, avg_samples: 128.0, ips: 120.17133 samples/s, eta: 0:42:46
[2022/08/25 01:27:49] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:28:12] ppocr INFO: epoch: [94/100], global_step: 5460, lr: 0.000038, acc: 0.960937, norm_edit_dis: 0.992006, loss: 0.252021, avg_reader_cost: 0.22799 s, avg_batch_cost: 1.29362 s, avg_samples: 128.0, ips: 98.94728 samples/s, eta: 0:42:26
[2022/08/25 01:28:33] ppocr INFO: epoch: [94/100], global_step: 5480, lr: 0.000038, acc: 0.960937, norm_edit_dis: 0.989715, loss: 0.269966, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06083 s, avg_samples: 128.0, ips: 120.66055 samples/s, eta: 0:42:04
[2022/08/25 01:28:55] ppocr INFO: epoch: [94/100], global_step: 5500, lr: 0.000037, acc: 0.960937, norm_edit_dis: 0.992452, loss: 0.238379, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06102 s, avg_samples: 128.0, ips: 120.63809 samples/s, eta: 0:41:43
[2022/08/25 01:29:16] ppocr INFO: epoch: [94/100], global_step: 5520, lr: 0.000037, acc: 0.964844, norm_edit_dis: 0.992041, loss: 0.250155, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06236 s, avg_samples: 128.0, ips: 120.48618 samples/s, eta: 0:41:21
[2022/08/25 01:29:37] ppocr INFO: epoch: [94/100], global_step: 5540, lr: 0.000037, acc: 0.960937, norm_edit_dis: 0.992935, loss: 0.223480, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06342 s, avg_samples: 128.0, ips: 120.36603 samples/s, eta: 0:40:59
[2022/08/25 01:29:58] ppocr INFO: epoch: [94/100], global_step: 5560, lr: 0.000036, acc: 0.960937, norm_edit_dis: 0.991450, loss: 0.281155, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06407 s, avg_samples: 128.0, ips: 120.29238 samples/s, eta: 0:40:38
[2022/08/25 01:30:20] ppocr INFO: epoch: [94/100], global_step: 5580, lr: 0.000036, acc: 0.949219, norm_edit_dis: 0.990225, loss: 0.344612, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06502 s, avg_samples: 128.0, ips: 120.18575 samples/s, eta: 0:40:16
[2022/08/25 01:30:41] ppocr INFO: epoch: [94/100], global_step: 5600, lr: 0.000036, acc: 0.957031, norm_edit_dis: 0.991194, loss: 0.229903, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06472 s, avg_samples: 128.0, ips: 120.21978 samples/s, eta: 0:39:54
[2022/08/25 01:31:02] ppocr INFO: epoch: [94/100], global_step: 5620, lr: 0.000035, acc: 0.960937, norm_edit_dis: 0.992812, loss: 0.272564, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06501 s, avg_samples: 128.0, ips: 120.18646 samples/s, eta: 0:39:33
[2022/08/25 01:31:24] ppocr INFO: epoch: [94/100], global_step: 5640, lr: 0.000035, acc: 0.953125, norm_edit_dis: 0.990368, loss: 0.296954, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06293 s, avg_samples: 128.0, ips: 120.42165 samples/s, eta: 0:39:11
[2022/08/25 01:31:45] ppocr INFO: epoch: [94/100], global_step: 5660, lr: 0.000035, acc: 0.960937, norm_edit_dis: 0.994015, loss: 0.273325, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06549 s, avg_samples: 128.0, ips: 120.13267 samples/s, eta: 0:38:49
[2022/08/25 01:32:06] ppocr INFO: epoch: [94/100], global_step: 5680, lr: 0.000034, acc: 0.960937, norm_edit_dis: 0.991225, loss: 0.270711, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06490 s, avg_samples: 128.0, ips: 120.19855 samples/s, eta: 0:38:28
[2022/08/25 01:32:28] ppocr INFO: epoch: [94/100], global_step: 5700, lr: 0.000034, acc: 0.960937, norm_edit_dis: 0.991811, loss: 0.276889, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06583 s, avg_samples: 128.0, ips: 120.09463 samples/s, eta: 0:38:06
[2022/08/25 01:32:49] ppocr INFO: epoch: [94/100], global_step: 5720, lr: 0.000034, acc: 0.964844, norm_edit_dis: 0.992039, loss: 0.290417, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06571 s, avg_samples: 128.0, ips: 120.10764 samples/s, eta: 0:37:44
[2022/08/25 01:33:10] ppocr INFO: epoch: [94/100], global_step: 5740, lr: 0.000033, acc: 0.953125, norm_edit_dis: 0.992751, loss: 0.326629, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06618 s, avg_samples: 128.0, ips: 120.05427 samples/s, eta: 0:37:23
[2022/08/25 01:33:32] ppocr INFO: epoch: [94/100], global_step: 5760, lr: 0.000033, acc: 0.964844, norm_edit_dis: 0.992330, loss: 0.299766, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06564 s, avg_samples: 128.0, ips: 120.11520 samples/s, eta: 0:37:01
[2022/08/25 01:33:53] ppocr INFO: epoch: [94/100], global_step: 5780, lr: 0.000033, acc: 0.960937, norm_edit_dis: 0.992964, loss: 0.281730, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06089 s, avg_samples: 128.0, ips: 120.65378 samples/s, eta: 0:36:39
[2022/08/25 01:33:56] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:34:19] ppocr INFO: epoch: [95/100], global_step: 5800, lr: 0.000032, acc: 0.964844, norm_edit_dis: 0.993627, loss: 0.291563, avg_reader_cost: 0.23079 s, avg_batch_cost: 1.29523 s, avg_samples: 128.0, ips: 98.82445 samples/s, eta: 0:36:19
[2022/08/25 01:34:40] ppocr INFO: epoch: [95/100], global_step: 5820, lr: 0.000032, acc: 0.968750, norm_edit_dis: 0.993012, loss: 0.340954, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06237 s, avg_samples: 128.0, ips: 120.48519 samples/s, eta: 0:35:57
[2022/08/25 01:35:01] ppocr INFO: epoch: [95/100], global_step: 5840, lr: 0.000032, acc: 0.957031, norm_edit_dis: 0.993720, loss: 0.295266, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06326 s, avg_samples: 128.0, ips: 120.38489 samples/s, eta: 0:35:36
[2022/08/25 01:35:23] ppocr INFO: epoch: [95/100], global_step: 5860, lr: 0.000031, acc: 0.949219, norm_edit_dis: 0.991943, loss: 0.260945, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06405 s, avg_samples: 128.0, ips: 120.29548 samples/s, eta: 0:35:14
[2022/08/25 01:35:44] ppocr INFO: epoch: [95/100], global_step: 5880, lr: 0.000031, acc: 0.964844, norm_edit_dis: 0.993040, loss: 0.295934, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06562 s, avg_samples: 128.0, ips: 120.11837 samples/s, eta: 0:34:52
[2022/08/25 01:36:05] ppocr INFO: epoch: [95/100], global_step: 5900, lr: 0.000031, acc: 0.960937, norm_edit_dis: 0.993320, loss: 0.287681, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06373 s, avg_samples: 128.0, ips: 120.33101 samples/s, eta: 0:34:31
[2022/08/25 01:36:26] ppocr INFO: epoch: [95/100], global_step: 5920, lr: 0.000030, acc: 0.960937, norm_edit_dis: 0.991697, loss: 0.305313, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06449 s, avg_samples: 128.0, ips: 120.24567 samples/s, eta: 0:34:09
[2022/08/25 01:36:48] ppocr INFO: epoch: [95/100], global_step: 5940, lr: 0.000030, acc: 0.957031, norm_edit_dis: 0.993248, loss: 0.278163, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06638 s, avg_samples: 128.0, ips: 120.03217 samples/s, eta: 0:33:47
[2022/08/25 01:37:09] ppocr INFO: epoch: [95/100], global_step: 5960, lr: 0.000030, acc: 0.964844, norm_edit_dis: 0.994265, loss: 0.262168, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06505 s, avg_samples: 128.0, ips: 120.18255 samples/s, eta: 0:33:26
[2022/08/25 01:37:30] ppocr INFO: epoch: [95/100], global_step: 5980, lr: 0.000029, acc: 0.960937, norm_edit_dis: 0.992080, loss: 0.292569, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06354 s, avg_samples: 128.0, ips: 120.35229 samples/s, eta: 0:33:04
[2022/08/25 01:37:52] ppocr INFO: epoch: [95/100], global_step: 6000, lr: 0.000029, acc: 0.960937, norm_edit_dis: 0.992606, loss: 0.269107, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06430 s, avg_samples: 128.0, ips: 120.26691 samples/s, eta: 0:32:42
eval model:: 100%|██████████████████████████████| 18/18 [00:07<00:00,  2.50it/s]
[2022/08/25 01:38:00] ppocr INFO: cur metric, acc: 0.9759930118900347, norm_edit_dis: 0.9888217300267031, fps: 440.0080657266238
[2022/08/25 01:38:02] ppocr INFO: save best model is to ./output/rec/PPOCRV3_0.5/best_accuracy
[2022/08/25 01:38:02] ppocr INFO: best metric, acc: 0.9759930118900347, norm_edit_dis: 0.9888217300267031, fps: 440.0080657266238, best_epoch: 95, start_epoch: 78
[2022/08/25 01:38:24] ppocr INFO: epoch: [95/100], global_step: 6020, lr: 0.000029, acc: 0.960937, norm_edit_dis: 0.993019, loss: 0.276715, avg_reader_cost: 0.00181 s, avg_batch_cost: 1.06507 s, avg_samples: 128.0, ips: 120.18030 samples/s, eta: 0:32:21
[2022/08/25 01:38:45] ppocr INFO: epoch: [95/100], global_step: 6040, lr: 0.000028, acc: 0.960937, norm_edit_dis: 0.991629, loss: 0.240549, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06360 s, avg_samples: 128.0, ips: 120.34556 samples/s, eta: 0:31:59
[2022/08/25 01:39:06] ppocr INFO: epoch: [95/100], global_step: 6060, lr: 0.000028, acc: 0.957031, norm_edit_dis: 0.992465, loss: 0.285926, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06358 s, avg_samples: 128.0, ips: 120.34771 samples/s, eta: 0:31:37
[2022/08/25 01:39:27] ppocr INFO: epoch: [95/100], global_step: 6080, lr: 0.000028, acc: 0.960937, norm_edit_dis: 0.993162, loss: 0.239741, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06466 s, avg_samples: 128.0, ips: 120.22583 samples/s, eta: 0:31:16
[2022/08/25 01:39:49] ppocr INFO: epoch: [95/100], global_step: 6100, lr: 0.000028, acc: 0.964844, norm_edit_dis: 0.993236, loss: 0.265314, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06614 s, avg_samples: 128.0, ips: 120.05967 samples/s, eta: 0:30:54
[2022/08/25 01:40:10] ppocr INFO: epoch: [95/100], global_step: 6120, lr: 0.000027, acc: 0.964844, norm_edit_dis: 0.992557, loss: 0.295307, avg_reader_cost: 0.00047 s, avg_batch_cost: 1.06710 s, avg_samples: 128.0, ips: 119.95158 samples/s, eta: 0:30:33
[2022/08/25 01:40:13] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:40:36] ppocr INFO: epoch: [96/100], global_step: 6140, lr: 0.000027, acc: 0.960937, norm_edit_dis: 0.991401, loss: 0.263386, avg_reader_cost: 0.22461 s, avg_batch_cost: 1.29670 s, avg_samples: 128.0, ips: 98.71233 samples/s, eta: 0:30:12
[2022/08/25 01:40:57] ppocr INFO: epoch: [96/100], global_step: 6160, lr: 0.000027, acc: 0.964844, norm_edit_dis: 0.992899, loss: 0.279949, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06560 s, avg_samples: 128.0, ips: 120.11962 samples/s, eta: 0:29:51
[2022/08/25 01:41:19] ppocr INFO: epoch: [96/100], global_step: 6180, lr: 0.000026, acc: 0.960937, norm_edit_dis: 0.990779, loss: 0.272377, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06699 s, avg_samples: 128.0, ips: 119.96349 samples/s, eta: 0:29:29
[2022/08/25 01:41:40] ppocr INFO: epoch: [96/100], global_step: 6200, lr: 0.000026, acc: 0.960937, norm_edit_dis: 0.993655, loss: 0.280393, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06642 s, avg_samples: 128.0, ips: 120.02821 samples/s, eta: 0:29:07
[2022/08/25 01:42:01] ppocr INFO: epoch: [96/100], global_step: 6220, lr: 0.000026, acc: 0.960937, norm_edit_dis: 0.993130, loss: 0.309715, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06738 s, avg_samples: 128.0, ips: 119.91987 samples/s, eta: 0:28:46
[2022/08/25 01:42:23] ppocr INFO: epoch: [96/100], global_step: 6240, lr: 0.000025, acc: 0.960937, norm_edit_dis: 0.992188, loss: 0.276981, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06608 s, avg_samples: 128.0, ips: 120.06605 samples/s, eta: 0:28:24
[2022/08/25 01:42:44] ppocr INFO: epoch: [96/100], global_step: 6260, lr: 0.000025, acc: 0.960937, norm_edit_dis: 0.992471, loss: 0.310961, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06667 s, avg_samples: 128.0, ips: 119.99929 samples/s, eta: 0:28:02
[2022/08/25 01:43:05] ppocr INFO: epoch: [96/100], global_step: 6280, lr: 0.000025, acc: 0.960937, norm_edit_dis: 0.992006, loss: 0.257836, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06438 s, avg_samples: 128.0, ips: 120.25725 samples/s, eta: 0:27:41
[2022/08/25 01:43:27] ppocr INFO: epoch: [96/100], global_step: 6300, lr: 0.000025, acc: 0.960937, norm_edit_dis: 0.992423, loss: 0.291288, avg_reader_cost: 0.00054 s, avg_batch_cost: 1.06214 s, avg_samples: 128.0, ips: 120.51106 samples/s, eta: 0:27:19
[2022/08/25 01:43:48] ppocr INFO: epoch: [96/100], global_step: 6320, lr: 0.000024, acc: 0.957031, norm_edit_dis: 0.992413, loss: 0.241745, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06607 s, avg_samples: 128.0, ips: 120.06759 samples/s, eta: 0:26:57
[2022/08/25 01:44:09] ppocr INFO: epoch: [96/100], global_step: 6340, lr: 0.000024, acc: 0.960937, norm_edit_dis: 0.992890, loss: 0.256064, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06600 s, avg_samples: 128.0, ips: 120.07458 samples/s, eta: 0:26:36
[2022/08/25 01:44:31] ppocr INFO: epoch: [96/100], global_step: 6360, lr: 0.000024, acc: 0.964844, norm_edit_dis: 0.991362, loss: 0.310378, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06575 s, avg_samples: 128.0, ips: 120.10362 samples/s, eta: 0:26:14
[2022/08/25 01:44:52] ppocr INFO: epoch: [96/100], global_step: 6380, lr: 0.000023, acc: 0.957031, norm_edit_dis: 0.991687, loss: 0.309480, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06465 s, avg_samples: 128.0, ips: 120.22691 samples/s, eta: 0:25:53
[2022/08/25 01:45:13] ppocr INFO: epoch: [96/100], global_step: 6400, lr: 0.000023, acc: 0.968750, norm_edit_dis: 0.993175, loss: 0.228975, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06639 s, avg_samples: 128.0, ips: 120.03134 samples/s, eta: 0:25:31
[2022/08/25 01:45:35] ppocr INFO: epoch: [96/100], global_step: 6420, lr: 0.000023, acc: 0.964844, norm_edit_dis: 0.993479, loss: 0.292460, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06496 s, avg_samples: 128.0, ips: 120.19272 samples/s, eta: 0:25:09
[2022/08/25 01:45:56] ppocr INFO: epoch: [96/100], global_step: 6440, lr: 0.000023, acc: 0.960937, norm_edit_dis: 0.994076, loss: 0.315330, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06354 s, avg_samples: 128.0, ips: 120.35227 samples/s, eta: 0:24:48
[2022/08/25 01:46:17] ppocr INFO: epoch: [96/100], global_step: 6460, lr: 0.000022, acc: 0.964844, norm_edit_dis: 0.992544, loss: 0.259433, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06329 s, avg_samples: 128.0, ips: 120.38127 samples/s, eta: 0:24:26
[2022/08/25 01:46:20] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:46:43] ppocr INFO: epoch: [97/100], global_step: 6480, lr: 0.000022, acc: 0.960937, norm_edit_dis: 0.992849, loss: 0.271306, avg_reader_cost: 0.22089 s, avg_batch_cost: 1.28646 s, avg_samples: 128.0, ips: 99.49781 samples/s, eta: 0:24:05
[2022/08/25 01:47:04] ppocr INFO: epoch: [97/100], global_step: 6500, lr: 0.000022, acc: 0.953125, norm_edit_dis: 0.990774, loss: 0.260268, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06460 s, avg_samples: 128.0, ips: 120.23272 samples/s, eta: 0:23:44
[2022/08/25 01:47:26] ppocr INFO: epoch: [97/100], global_step: 6520, lr: 0.000022, acc: 0.968750, norm_edit_dis: 0.994140, loss: 0.237290, avg_reader_cost: 0.00051 s, avg_batch_cost: 1.06487 s, avg_samples: 128.0, ips: 120.20235 samples/s, eta: 0:23:22
[2022/08/25 01:47:47] ppocr INFO: epoch: [97/100], global_step: 6540, lr: 0.000021, acc: 0.957031, norm_edit_dis: 0.992153, loss: 0.247355, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06466 s, avg_samples: 128.0, ips: 120.22631 samples/s, eta: 0:23:00
[2022/08/25 01:48:08] ppocr INFO: epoch: [97/100], global_step: 6560, lr: 0.000021, acc: 0.953125, norm_edit_dis: 0.991194, loss: 0.287906, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06454 s, avg_samples: 128.0, ips: 120.23985 samples/s, eta: 0:22:39
[2022/08/25 01:48:29] ppocr INFO: epoch: [97/100], global_step: 6580, lr: 0.000021, acc: 0.960937, norm_edit_dis: 0.994546, loss: 0.194431, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06595 s, avg_samples: 128.0, ips: 120.08102 samples/s, eta: 0:22:17
[2022/08/25 01:48:51] ppocr INFO: epoch: [97/100], global_step: 6600, lr: 0.000021, acc: 0.964844, norm_edit_dis: 0.993144, loss: 0.265878, avg_reader_cost: 0.00051 s, avg_batch_cost: 1.06442 s, avg_samples: 128.0, ips: 120.25372 samples/s, eta: 0:21:56
[2022/08/25 01:49:12] ppocr INFO: epoch: [97/100], global_step: 6620, lr: 0.000020, acc: 0.960937, norm_edit_dis: 0.990812, loss: 0.336426, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06360 s, avg_samples: 128.0, ips: 120.34561 samples/s, eta: 0:21:34
[2022/08/25 01:49:33] ppocr INFO: epoch: [97/100], global_step: 6640, lr: 0.000020, acc: 0.968750, norm_edit_dis: 0.993171, loss: 0.267464, avg_reader_cost: 0.00053 s, avg_batch_cost: 1.06508 s, avg_samples: 128.0, ips: 120.17863 samples/s, eta: 0:21:12
[2022/08/25 01:49:55] ppocr INFO: epoch: [97/100], global_step: 6660, lr: 0.000020, acc: 0.968750, norm_edit_dis: 0.993814, loss: 0.274757, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06576 s, avg_samples: 128.0, ips: 120.10167 samples/s, eta: 0:20:51
[2022/08/25 01:50:16] ppocr INFO: epoch: [97/100], global_step: 6680, lr: 0.000019, acc: 0.968750, norm_edit_dis: 0.995382, loss: 0.239265, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06534 s, avg_samples: 128.0, ips: 120.14890 samples/s, eta: 0:20:29
[2022/08/25 01:50:37] ppocr INFO: epoch: [97/100], global_step: 6700, lr: 0.000019, acc: 0.960937, norm_edit_dis: 0.993161, loss: 0.244231, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06458 s, avg_samples: 128.0, ips: 120.23539 samples/s, eta: 0:20:07
[2022/08/25 01:50:59] ppocr INFO: epoch: [97/100], global_step: 6720, lr: 0.000019, acc: 0.960937, norm_edit_dis: 0.993105, loss: 0.264068, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06546 s, avg_samples: 128.0, ips: 120.13558 samples/s, eta: 0:19:46
[2022/08/25 01:51:20] ppocr INFO: epoch: [97/100], global_step: 6740, lr: 0.000019, acc: 0.964844, norm_edit_dis: 0.993596, loss: 0.236834, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06638 s, avg_samples: 128.0, ips: 120.03188 samples/s, eta: 0:19:24
[2022/08/25 01:51:41] ppocr INFO: epoch: [97/100], global_step: 6760, lr: 0.000018, acc: 0.957031, norm_edit_dis: 0.991385, loss: 0.318493, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06434 s, avg_samples: 128.0, ips: 120.26194 samples/s, eta: 0:19:03
[2022/08/25 01:52:03] ppocr INFO: epoch: [97/100], global_step: 6780, lr: 0.000018, acc: 0.968750, norm_edit_dis: 0.994015, loss: 0.232673, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06559 s, avg_samples: 128.0, ips: 120.12173 samples/s, eta: 0:18:41
[2022/08/25 01:52:24] ppocr INFO: epoch: [97/100], global_step: 6800, lr: 0.000018, acc: 0.960937, norm_edit_dis: 0.992017, loss: 0.249960, avg_reader_cost: 0.00046 s, avg_batch_cost: 1.06665 s, avg_samples: 128.0, ips: 120.00200 samples/s, eta: 0:18:19
[2022/08/25 01:52:27] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:52:50] ppocr INFO: epoch: [98/100], global_step: 6820, lr: 0.000018, acc: 0.960937, norm_edit_dis: 0.991160, loss: 0.285255, avg_reader_cost: 0.23322 s, avg_batch_cost: 1.30371 s, avg_samples: 128.0, ips: 98.18158 samples/s, eta: 0:17:58
[2022/08/25 01:53:11] ppocr INFO: epoch: [98/100], global_step: 6840, lr: 0.000017, acc: 0.968750, norm_edit_dis: 0.993449, loss: 0.233170, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06717 s, avg_samples: 128.0, ips: 119.94334 samples/s, eta: 0:17:37
[2022/08/25 01:53:33] ppocr INFO: epoch: [98/100], global_step: 6860, lr: 0.000017, acc: 0.972656, norm_edit_dis: 0.995646, loss: 0.191811, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06837 s, avg_samples: 128.0, ips: 119.80839 samples/s, eta: 0:17:15
[2022/08/25 01:53:54] ppocr INFO: epoch: [98/100], global_step: 6880, lr: 0.000017, acc: 0.964844, norm_edit_dis: 0.993926, loss: 0.255197, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06611 s, avg_samples: 128.0, ips: 120.06287 samples/s, eta: 0:16:54
[2022/08/25 01:54:15] ppocr INFO: epoch: [98/100], global_step: 6900, lr: 0.000017, acc: 0.964844, norm_edit_dis: 0.992517, loss: 0.316043, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06567 s, avg_samples: 128.0, ips: 120.11257 samples/s, eta: 0:16:32
[2022/08/25 01:54:37] ppocr INFO: epoch: [98/100], global_step: 6920, lr: 0.000017, acc: 0.960937, norm_edit_dis: 0.993204, loss: 0.249734, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06450 s, avg_samples: 128.0, ips: 120.24448 samples/s, eta: 0:16:10
[2022/08/25 01:54:58] ppocr INFO: epoch: [98/100], global_step: 6940, lr: 0.000016, acc: 0.957031, norm_edit_dis: 0.992268, loss: 0.282014, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06207 s, avg_samples: 128.0, ips: 120.51911 samples/s, eta: 0:15:49
[2022/08/25 01:55:19] ppocr INFO: epoch: [98/100], global_step: 6960, lr: 0.000016, acc: 0.968750, norm_edit_dis: 0.993880, loss: 0.194212, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06265 s, avg_samples: 128.0, ips: 120.45388 samples/s, eta: 0:15:27
[2022/08/25 01:55:40] ppocr INFO: epoch: [98/100], global_step: 6980, lr: 0.000016, acc: 0.960937, norm_edit_dis: 0.993064, loss: 0.218887, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06318 s, avg_samples: 128.0, ips: 120.39376 samples/s, eta: 0:15:06
[2022/08/25 01:56:02] ppocr INFO: epoch: [98/100], global_step: 7000, lr: 0.000016, acc: 0.960937, norm_edit_dis: 0.991858, loss: 0.307710, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06611 s, avg_samples: 128.0, ips: 120.06259 samples/s, eta: 0:14:44
[2022/08/25 01:56:23] ppocr INFO: epoch: [98/100], global_step: 7020, lr: 0.000015, acc: 0.972656, norm_edit_dis: 0.995499, loss: 0.200291, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06579 s, avg_samples: 128.0, ips: 120.09829 samples/s, eta: 0:14:22
[2022/08/25 01:56:44] ppocr INFO: epoch: [98/100], global_step: 7040, lr: 0.000015, acc: 0.960937, norm_edit_dis: 0.995495, loss: 0.200220, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06505 s, avg_samples: 128.0, ips: 120.18162 samples/s, eta: 0:14:01
[2022/08/25 01:57:06] ppocr INFO: epoch: [98/100], global_step: 7060, lr: 0.000015, acc: 0.960937, norm_edit_dis: 0.993354, loss: 0.269212, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06626 s, avg_samples: 128.0, ips: 120.04595 samples/s, eta: 0:13:39
[2022/08/25 01:57:27] ppocr INFO: epoch: [98/100], global_step: 7080, lr: 0.000015, acc: 0.968750, norm_edit_dis: 0.994959, loss: 0.246107, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06703 s, avg_samples: 128.0, ips: 119.95867 samples/s, eta: 0:13:18
[2022/08/25 01:57:48] ppocr INFO: epoch: [98/100], global_step: 7100, lr: 0.000014, acc: 0.960937, norm_edit_dis: 0.994062, loss: 0.259850, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06597 s, avg_samples: 128.0, ips: 120.07863 samples/s, eta: 0:12:56
[2022/08/25 01:58:10] ppocr INFO: epoch: [98/100], global_step: 7120, lr: 0.000014, acc: 0.968750, norm_edit_dis: 0.992867, loss: 0.218677, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06571 s, avg_samples: 128.0, ips: 120.10824 samples/s, eta: 0:12:34
[2022/08/25 01:58:31] ppocr INFO: epoch: [98/100], global_step: 7140, lr: 0.000014, acc: 0.957031, norm_edit_dis: 0.991765, loss: 0.249263, avg_reader_cost: 0.00056 s, avg_batch_cost: 1.06181 s, avg_samples: 128.0, ips: 120.54854 samples/s, eta: 0:12:13
[2022/08/25 01:58:34] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 01:58:57] ppocr INFO: epoch: [99/100], global_step: 7160, lr: 0.000014, acc: 0.953125, norm_edit_dis: 0.988960, loss: 0.407955, avg_reader_cost: 0.22736 s, avg_batch_cost: 1.29111 s, avg_samples: 128.0, ips: 99.13939 samples/s, eta: 0:11:52
[2022/08/25 01:59:18] ppocr INFO: epoch: [99/100], global_step: 7180, lr: 0.000014, acc: 0.964844, norm_edit_dis: 0.994590, loss: 0.191439, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06280 s, avg_samples: 128.0, ips: 120.43648 samples/s, eta: 0:11:30
[2022/08/25 01:59:39] ppocr INFO: epoch: [99/100], global_step: 7200, lr: 0.000013, acc: 0.964844, norm_edit_dis: 0.994269, loss: 0.212174, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06291 s, avg_samples: 128.0, ips: 120.42400 samples/s, eta: 0:11:08
[2022/08/25 02:00:01] ppocr INFO: epoch: [99/100], global_step: 7220, lr: 0.000013, acc: 0.968750, norm_edit_dis: 0.993088, loss: 0.253993, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06287 s, avg_samples: 128.0, ips: 120.42823 samples/s, eta: 0:10:47
[2022/08/25 02:00:22] ppocr INFO: epoch: [99/100], global_step: 7240, lr: 0.000013, acc: 0.968750, norm_edit_dis: 0.993079, loss: 0.236398, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06231 s, avg_samples: 128.0, ips: 120.49212 samples/s, eta: 0:10:25
[2022/08/25 02:00:43] ppocr INFO: epoch: [99/100], global_step: 7260, lr: 0.000013, acc: 0.964844, norm_edit_dis: 0.993235, loss: 0.260110, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06280 s, avg_samples: 128.0, ips: 120.43709 samples/s, eta: 0:10:04
[2022/08/25 02:01:04] ppocr INFO: epoch: [99/100], global_step: 7280, lr: 0.000013, acc: 0.968750, norm_edit_dis: 0.994878, loss: 0.202151, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06285 s, avg_samples: 128.0, ips: 120.43061 samples/s, eta: 0:09:42
[2022/08/25 02:01:26] ppocr INFO: epoch: [99/100], global_step: 7300, lr: 0.000012, acc: 0.968750, norm_edit_dis: 0.994239, loss: 0.225898, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06219 s, avg_samples: 128.0, ips: 120.50542 samples/s, eta: 0:09:20
[2022/08/25 02:01:47] ppocr INFO: epoch: [99/100], global_step: 7320, lr: 0.000012, acc: 0.968750, norm_edit_dis: 0.994815, loss: 0.298698, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06264 s, avg_samples: 128.0, ips: 120.45498 samples/s, eta: 0:08:59
[2022/08/25 02:02:08] ppocr INFO: epoch: [99/100], global_step: 7340, lr: 0.000012, acc: 0.960937, norm_edit_dis: 0.993930, loss: 0.190748, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06328 s, avg_samples: 128.0, ips: 120.38220 samples/s, eta: 0:08:37
[2022/08/25 02:02:29] ppocr INFO: epoch: [99/100], global_step: 7360, lr: 0.000012, acc: 0.968750, norm_edit_dis: 0.992948, loss: 0.282006, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06304 s, avg_samples: 128.0, ips: 120.40970 samples/s, eta: 0:08:16
[2022/08/25 02:02:51] ppocr INFO: epoch: [99/100], global_step: 7380, lr: 0.000012, acc: 0.968750, norm_edit_dis: 0.994220, loss: 0.225243, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06406 s, avg_samples: 128.0, ips: 120.29384 samples/s, eta: 0:07:54
[2022/08/25 02:03:12] ppocr INFO: epoch: [99/100], global_step: 7400, lr: 0.000011, acc: 0.968750, norm_edit_dis: 0.993263, loss: 0.269360, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06592 s, avg_samples: 128.0, ips: 120.08435 samples/s, eta: 0:07:32
[2022/08/25 02:03:33] ppocr INFO: epoch: [99/100], global_step: 7420, lr: 0.000011, acc: 0.964844, norm_edit_dis: 0.993949, loss: 0.205838, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06535 s, avg_samples: 128.0, ips: 120.14778 samples/s, eta: 0:07:11
[2022/08/25 02:03:55] ppocr INFO: epoch: [99/100], global_step: 7440, lr: 0.000011, acc: 0.960937, norm_edit_dis: 0.993421, loss: 0.265975, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06609 s, avg_samples: 128.0, ips: 120.06537 samples/s, eta: 0:06:49
[2022/08/25 02:04:16] ppocr INFO: epoch: [99/100], global_step: 7460, lr: 0.000011, acc: 0.964844, norm_edit_dis: 0.993491, loss: 0.266176, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06474 s, avg_samples: 128.0, ips: 120.21665 samples/s, eta: 0:06:28
[2022/08/25 02:04:37] ppocr INFO: epoch: [99/100], global_step: 7480, lr: 0.000011, acc: 0.953125, norm_edit_dis: 0.990372, loss: 0.350108, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06308 s, avg_samples: 128.0, ips: 120.40518 samples/s, eta: 0:06:06
[2022/08/25 02:04:40] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 02:05:03] ppocr INFO: epoch: [100/100], global_step: 7500, lr: 0.000010, acc: 0.968750, norm_edit_dis: 0.993258, loss: 0.320832, avg_reader_cost: 0.21639 s, avg_batch_cost: 1.28145 s, avg_samples: 128.0, ips: 99.88670 samples/s, eta: 0:05:45
[2022/08/25 02:05:24] ppocr INFO: epoch: [100/100], global_step: 7520, lr: 0.000010, acc: 0.968750, norm_edit_dis: 0.993857, loss: 0.249691, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06448 s, avg_samples: 128.0, ips: 120.24599 samples/s, eta: 0:05:23
[2022/08/25 02:05:46] ppocr INFO: epoch: [100/100], global_step: 7540, lr: 0.000010, acc: 0.964844, norm_edit_dis: 0.994093, loss: 0.224872, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06518 s, avg_samples: 128.0, ips: 120.16775 samples/s, eta: 0:05:02
[2022/08/25 02:06:07] ppocr INFO: epoch: [100/100], global_step: 7560, lr: 0.000010, acc: 0.960937, norm_edit_dis: 0.990818, loss: 0.278519, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06601 s, avg_samples: 128.0, ips: 120.07392 samples/s, eta: 0:04:40
[2022/08/25 02:06:28] ppocr INFO: epoch: [100/100], global_step: 7580, lr: 0.000010, acc: 0.957031, norm_edit_dis: 0.992707, loss: 0.249093, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06570 s, avg_samples: 128.0, ips: 120.10861 samples/s, eta: 0:04:18
[2022/08/25 02:06:50] ppocr INFO: epoch: [100/100], global_step: 7600, lr: 0.000009, acc: 0.953125, norm_edit_dis: 0.992663, loss: 0.298814, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06712 s, avg_samples: 128.0, ips: 119.94849 samples/s, eta: 0:03:57
[2022/08/25 02:07:11] ppocr INFO: epoch: [100/100], global_step: 7620, lr: 0.000009, acc: 0.968750, norm_edit_dis: 0.993274, loss: 0.268467, avg_reader_cost: 0.00052 s, avg_batch_cost: 1.06683 s, avg_samples: 128.0, ips: 119.98142 samples/s, eta: 0:03:35
[2022/08/25 02:07:32] ppocr INFO: epoch: [100/100], global_step: 7640, lr: 0.000009, acc: 0.960937, norm_edit_dis: 0.993262, loss: 0.221667, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06437 s, avg_samples: 128.0, ips: 120.25940 samples/s, eta: 0:03:14
[2022/08/25 02:07:53] ppocr INFO: epoch: [100/100], global_step: 7660, lr: 0.000009, acc: 0.968750, norm_edit_dis: 0.994456, loss: 0.249594, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06383 s, avg_samples: 128.0, ips: 120.31969 samples/s, eta: 0:02:52
[2022/08/25 02:08:15] ppocr INFO: epoch: [100/100], global_step: 7680, lr: 0.000009, acc: 0.960937, norm_edit_dis: 0.990408, loss: 0.282642, avg_reader_cost: 0.00043 s, avg_batch_cost: 1.06582 s, avg_samples: 128.0, ips: 120.09559 samples/s, eta: 0:02:30
[2022/08/25 02:08:36] ppocr INFO: epoch: [100/100], global_step: 7700, lr: 0.000009, acc: 0.960937, norm_edit_dis: 0.992553, loss: 0.325926, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06579 s, avg_samples: 128.0, ips: 120.09849 samples/s, eta: 0:02:09
[2022/08/25 02:08:57] ppocr INFO: epoch: [100/100], global_step: 7720, lr: 0.000008, acc: 0.960937, norm_edit_dis: 0.992186, loss: 0.313792, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06607 s, avg_samples: 128.0, ips: 120.06693 samples/s, eta: 0:01:47
[2022/08/25 02:09:19] ppocr INFO: epoch: [100/100], global_step: 7740, lr: 0.000008, acc: 0.953125, norm_edit_dis: 0.990970, loss: 0.296407, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06574 s, avg_samples: 128.0, ips: 120.10400 samples/s, eta: 0:01:26
[2022/08/25 02:09:40] ppocr INFO: epoch: [100/100], global_step: 7760, lr: 0.000008, acc: 0.968750, norm_edit_dis: 0.994286, loss: 0.217486, avg_reader_cost: 0.00041 s, avg_batch_cost: 1.06503 s, avg_samples: 128.0, ips: 120.18461 samples/s, eta: 0:01:04
[2022/08/25 02:10:01] ppocr INFO: epoch: [100/100], global_step: 7780, lr: 0.000008, acc: 0.968750, norm_edit_dis: 0.993447, loss: 0.244423, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06611 s, avg_samples: 128.0, ips: 120.06277 samples/s, eta: 0:00:43
[2022/08/25 02:10:23] ppocr INFO: epoch: [100/100], global_step: 7800, lr: 0.000008, acc: 0.976562, norm_edit_dis: 0.995224, loss: 0.205109, avg_reader_cost: 0.00042 s, avg_batch_cost: 1.06649 s, avg_samples: 128.0, ips: 120.02025 samples/s, eta: 0:00:21
[2022/08/25 02:10:44] ppocr INFO: epoch: [100/100], global_step: 7820, lr: 0.000008, acc: 0.968750, norm_edit_dis: 0.993583, loss: 0.215131, avg_reader_cost: 0.00045 s, avg_batch_cost: 1.06601 s, avg_samples: 128.0, ips: 120.07396 samples/s, eta: 0:00:00
[2022/08/25 02:10:47] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/latest
[2022/08/25 02:10:48] ppocr INFO: save model in ./output/rec/PPOCRV3_0.5/iter_epoch_100
[2022/08/25 02:10:48] ppocr INFO: best metric, acc: 0.9759930118900347, norm_edit_dis: 0.9888217300267031, fps: 440.0080657266238, best_epoch: 95, start_epoch: 78

3.3 模型预测

%cd ./PaddleOCR

/home/aistudio/PaddleOCR

! python3 tools/infer_rec.py -c ./configs/rec/PP-OCRv3/en_PP-OCRv3_rec.yml -o Global.checkpoints=../output/rec/PPOCRV3_0.5/best_accuracy

[2022/08/25 08:12:24] ppocr INFO: success!

# 生成提交结果
data = pd.read_csv('./output/rec/predicts_ppocrv3_0.5_en.txt',encoding='UTF-8',sep='\t',header=None)
with open('../pdmodel_result_infer.txt', 'w', encoding='UTF-8') as f:
    f.write('new_name'+'\t'+'value'+'\n')
    for i in range(len(data)):
        name,label,score = data.iloc[i,:]
        text = name.replace('/home/aistudio/data/test_images/','')+'\t'+str(label)
        f.write(text+'\n')

3.4 模型部署

模型部署阶段，需要将训练好的模型转为静态再使用onnx进行推理

导出模型
转为onnx

%cd ./PaddleOCR
!python3 ./tools/export_model.py \
    -o Global.pretrained_model= home/aistudio/output/rec/PPOCRV3_0.5/best_accuracy\
    -o Global.save_inference_dir=../inference\
    -c ./configs/rec/PP-OCRv3/en_PP-OCRv3_rec.yml \

[Errno 2] No such file or directory: './PaddleOCR'
/home/aistudio/PaddleOCR
W0825 08:16:31.621683 15845 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0825 08:16:31.626344 15845 gpu_resources.cc:91] device: 0, cuDNN Version: 7.6.
[2022/08/25 08:16:32] ppocr INFO: train from scratch
[2022/08/25 08:16:35] ppocr INFO: inference model is saved to ../inference/inference

# 将best_accuracy模型及log复制到work保存
!cp /home/aistudio/output/rec/PPOCRV3_0.5/best_accuracy.pdparams /home/aistudio/work/best_accuracy.pdparams
!cp /home/aistudio/output/rec/PPOCRV3_0.5/best_accuracy.states /home/aistudio/work/best_accuracy.states
!cp /home/aistudio/output/rec/PPOCRV3_0.5/best_accuracy.pdopt /home/aistudio/work/best_accuracy.pdopt
!cp /home/aistudio/output/rec/PPOCRV3_0.5/train.log /home/aistudio/work/train.log

# 导出onnx模型
%cd ~
!paddle2onnx --model_dir inference \
            --model_filename inference.pdmodel \
            --params_filename inference.pdiparams \
            --save_file work/model.onnx \
            --enable_dev_version True \
arams_filename inference.pdiparams \
            --save_file work/model.onnx \
            --enable_dev_version True \
            --opset_version=11

/home/aistudio
[Paddle2ONNX] Start to parse PaddlePaddle model...
[Paddle2ONNX] Model file path: inference/inference.pdmodel
[Paddle2ONNX] Paramters file path: inference/inference.pdiparams
[Paddle2ONNX] Start to parsing Paddle model...
[Paddle2ONNX] Use opset_version = 11 for ONNX export.
[Paddle2ONNX] PaddlePaddle model is exported as ONNX format now.
2022-08-25 08:16:48 [INFO]	===============Make PaddlePaddle Better!================
2022-08-25 08:16:48 [INFO]	A little survey: https://iwenjuan.baidu.com/?code=r8hu2s

四、总结

本项目主要实现SVTR模型的训练，验证及onnx导出。
本项目主要删除了模型的绝对位置编码及修改推理部分代码，对于使用相对位置编码更换则会在之后实现。

test指标即为飞桨学习赛：中文场景文字识别提交结果

Epoch	Lr	batchSize	Transform	Backbone	char_num	Neck	Head	Loss	transforms	Train Time	Train Ips	train	eval	test	备注
100	0.001	128	None	SVTR-T	25	SeEncoder(reshape)	CTCHead	CTCLoss	None	3h55m	310	79	60.01	60.96	Normlize(ImageNet)
100	0.001	128	None	SVTR-T	25	SeeEncoder(reshape)	CTCHead	CTCLoss	None	4h5m	312	79	59.27	60.96	Normlize(0.5)
100	0.001	128	None	ppocrv3	-	svtr	MultiHead	MultiLoss	None	4h35m	274	50	45	48.21	scale(0.5)
100	0.001	128	None	ppocrv3	-	svtr	MultiHead	MultiLoss	None	6h28m	192	69.99	48.95	53.86	scale(1)
100	0.001	128	None	ppocrv3	-	svtr	MultiHead	MultiLoss	None	11h50m	110	83.59	55.95	58.72	scale(2)
100	0.001	128	None	ResNet	-	SeEncoder(rnn)	CTCHead	CTCHead	None	10h20m	125	96.8	97.5	63.72	ResNet34

注意:建议大家前往SVTR： Scene Text Recognition with a Single Visual Model查看原文

有任何问题，欢迎评论区留言交流。

此文章为搬运
原项目链接

百度飞桨AI Studio社区

学大模型，用大模型上飞桨星河社区！每天8点V100G算力免费领！免费领取ERNIE 4.0 100w Token >>>

更多推荐

利用Amazon Bedrock畅玩Claude 3等多种领先模型，抢占AI高地(体验倒计时4小时)

百度飞桨星河社区

RAPTOR：索引树状 RAG，使用树结构来捕捉文本的高级和低级细节

百度飞桨星河社区

MultiHop-RAG：多跳查询的基准检索增强生成

百度飞桨星河社区

所有评论(0)

查看更多评论

AI Studio

@m0_63642362

已为社区贡献1437条内容