基于PaddleOCR2.4的【常规赛：中文场景文字识别】Baseline
基于PaddleOCR2.4 中文场景文字识别，对比赛提供的数据进行预处理，选择CRNN网络，加载预训练模型进行快速训练并预测，从而快速实现中文场景文字识别。
·
一、常规赛：中文场景文字识别
比赛地址：https://aistudio.baidu.com/aistudio/competition/detail/20/0/datasets
1.比赛简介
中文场景文字识别技术在人们的日常生活中受到广泛关注，具有丰富的应用场景，如：拍照翻译、图像检索、场景理解等。然而，中文场景中的文字面临着包括光照变化、低分辨率、字体以及排布多样性、中文字符种类多等复杂情况。如何解决上述问题成为一项极具挑战性的任务。
中文场景文字识别常规赛全新升级，提供轻量级中文场景文字识别数据，要求选手使用飞桨框架，对图像区域中的文字行进行预测，并返回文字行的内容。
2.数据集描述
本次赛题数据集共包括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。
具体数据介绍
数据集中所有图像都经过一些预处理，如下图所示：
(a) 标注：久斯台球会所
(b) 标注：上海创科泵业制造有限公司
标注文件
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：
| name | value |
|---|---|
| 0.jpg | 文本0 |
| -------- | -------- |
| 1.jpg | 文本0 |
二、环境设置
PaddleOCR https://github.com/paddlepaddle/PaddleOCR 是一款全宇宙最强的用的OCR工具库，开箱即用，速度杠杠的。
# 从gitee上下载PaddleOCR代码，也可以从GitHub链接下载
!git clone https://gitee.com/paddlepaddle/PaddleOCR.git --depth=1
# 升级pip
!pip install -U pip
# 安装依赖
%cd ~/PaddleOCR
%pip install -r requirements.txt
%cd ~/PaddleOCR/
!tree -L 1
/home/aistudio/PaddleOCR
.
├── benchmark
├── configs
├── deploy
├── doc
├── __init__.py
├── LICENSE
├── MANIFEST.in
├── paddleocr.py
├── ppocr
├── PPOCRLabel
├── ppstructure
├── README_ch.md
├── README.md
├── requirements.txt
├── setup.py
├── StyleText
├── test_tipc
├── tools
└── train.sh
10 directories, 9 files
三、数据准备
据悉train数据集共10万张，解压，并划分出10000张作为测试集。
1.数据下载解压
# 解压缩数据集
%cd ~
!unzip -qa data/data62842/train_images.zip -d data/data62842/
!unzip -qa data/data62843/test_images.zip -d data/data62843/
/home/aistudio
# 使用命令查看训练数据文件夹下数据量是否是5万张
!cd ~/data/data62842/train_images && ls -l | grep "^-" | wc -l
50000
# 使用命令查看test数据文件夹下数据量是否是1万张
!cd ~/data/data62843/test_images && ls -l | grep "^-" | wc -l
10000
2. 数据集划分
# 读取数据列表文件
import pandas as pd
%cd ~
data_label=pd.read_csv('data/data62842/train_label.csv', encoding='gb2312')
data_label.head()
/home/aistudio
| name | value | |
|---|---|---|
| 0 | 0.jpg | 拉拉 |
| 1 | 1.jpg | ６号 |
| 2 | 2.jpg | 胖胖 |
| 3 | 3.jpg | 前门大栅栏总店 |
| 4 | 4.jpg | 你来就是旺季 |
# 对数据列表文件进行划分
%cd ~/data/data62842/
print(data_label.shape)
train=data_label[:45000]
val=data_label[45000:]
train.to_csv('train.txt',sep='\t',header=None,index=None)
val.to_csv('val.txt',sep='\t',header=None,index=None)
/home/aistudio/data/data62842
(50000, 2)
# 查看数量
print(train.shape)
print(val.shape)
(45000, 2)
(5000, 2)
!head val.txt
45000.jpg 责任单位：北京市环清环卫设施维修
45001.jpg 眼镜
45002.jpg 光临
45003.jpg 主治
45004.jpg 菜饭骨头汤
45005.jpg 理
45006.jpg 要多者提前预定
45007.jpg 干洗湿洗
45008.jpg 画布咖啡
45009.jpg 电焊、气割、专业自卸车
!head train.txt
0.jpg 拉拉
1.jpg ６号
2.jpg 胖胖
3.jpg 前门大栅栏总店
4.jpg 你来就是旺季
5.jpg 毛衣厂家直销
6.jpg 13761916218
7.jpg 福鼎白茶
8.jpg 妍心美容
9.jpg 童车童床
四、配置训练参数
以PaddleOCR/configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml为基准进行配置
1.配置模型网络
使用CRNN算法，backbone是MobileNetV3，损失函数是CTCLoss
Architecture:
model_type: rec
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: small
small_stride: [1, 2, 2, 2]
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 48
Head:
name: CTCHead
fc_decay: 0.00001
2.配置数据
对Train.data_dir, Train.label_file_list, Eval.data_dir, Eval.label_file_list进行配置
Train:
dataset:
name: SimpleDataSet
data_dir: /home/aistudio/data/data62842/train_images
label_file_list: ["/home/aistudio/data/data62842/train.txt"]
...
...
Eval:
dataset:
name: SimpleDataSet
data_dir: /home/aistudio/data/data62842/train_images
label_file_list: ["/home/aistudio/data/data62842/val.txt"]
3. 显卡、评估设置
use_gpu、cal_metric_during_train分别是GPU、评估开关
Global:
use_gpu: false # true 使用GPU
.....
cal_metric_during_train: False # true 打开评估
4. 多线程任务
Train.loader.num_workers:4
Eval.loader.num_workers: 4
5.完整配置
Global:
use_gpu: true
epoch_num: 500
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/rec_chinese_lite_v2.0
save_epoch_step: 3
# evaluation is run every 5000 iterations after the 4000th iteration
eval_batch_step: [0, 2000]
cal_metric_during_train: True
pretrained_model: ./ch_ppocr_mobile_v2.0_rec_pre/best_accuracy
checkpoints:
save_inference_dir:
use_visualdl: True
infer_img: doc/imgs_words/ch/word_1.jpg
# for data or label process
character_dict_path: ppocr/utils/ppocr_keys_v1.txt
max_text_length: 25
infer_mode: False
use_space_char: True
save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Cosine
learning_rate: 0.001
warmup_epoch: 5
regularizer:
name: 'L2'
factor: 0.00001
Architecture:
model_type: rec
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: small
small_stride: [1, 2, 2, 2]
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 48
Head:
name: CTCHead
fc_decay: 0.00001
Loss:
name: CTCLoss
PostProcess:
name: CTCLabelDecode
Metric:
name: RecMetric
main_indicator: acc
Train:
dataset:
name: SimpleDataSet
data_dir: /home/aistudio/data/data62842/train_images
label_file_list: ["/home/aistudio/data/data62842/train.txt"]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- RecAug:
- CTCLabelEncode: # Class handling label
- RecResizeImg:
image_shape: [3, 32, 320]
- KeepKeys:
keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
loader:
shuffle: True
batch_size_per_card: 256
drop_last: True
num_workers: 8
Eval:
dataset:
name: SimpleDataSet
data_dir: /home/aistudio/data/data62842/train_images
label_file_list: ["/home/aistudio/data/data62842/val.txt"]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- CTCLabelEncode: # Class handling label
- RecResizeImg:
image_shape: [3, 32, 320]
- KeepKeys:
keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
loader:
shuffle: False
drop_last: False
batch_size_per_card: 256
num_workers: 8
# 覆盖配置
!cp -f ~/rec_chinese_lite_train_v2.0.yml ~/PaddleOCR/configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml
cp: cannot stat '/home/aistudio/PaddleOCR/configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml': No such file or directory
6.使用预训练模型
据悉使用预训练模型，训练速度更快！！！
PaddleOCR提供的可下载模型包括推理模型、训练模型、预训练模型、slim模型，模型区别说明如下：
| 模型类型 | 模型格式 | 简介 |
|---|---|---|
| 推理模型 | inference.pdmodel、inference.pdiparams | 用于预测引擎推理，详情 |
| 训练模型、预训练模型 | *.pdparams、*.pdopt、*.states | 训练过程中保存的模型的参数、优化器状态和训练中间信息，多用于模型指标评估和恢复训练 |
| slim模型 | *.nb | 经过飞桨模型压缩工具PaddleSlim压缩后的模型，适用于移动端/IoT端等端侧部署场景（需使用飞桨Paddle Lite部署）。 |
各个模型的关系如下面的示意图所示。
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-gTxACxiT-1641602379723)(…/imgs/model_prod_flow_ch.png)]
文本检测模型
| 模型名称 | 模型简介 | 配置文件 | 推理模型大小 | 下载地址 |
|---|---|---|---|---|
| ch_ppocr_mobile_slim_v2.0_det | slim裁剪版超轻量模型，支持中英文、多语种文本检测 | ch_det_mv3_db_v2.0.yml | 2.6M | 推理模型 |
| ch_ppocr_mobile_v2.0_det | 原始超轻量模型，支持中英文、多语种文本检测 | ch_det_mv3_db_v2.0.yml | 3M | 推理模型 / 训练模型 |
| ch_ppocr_server_v2.0_det | 通用模型，支持中英文、多语种文本检测，比超轻量模型更大，但效果更好 | ch_det_res18_db_v2.0.yml | 47M | 推理模型 / 训练模型 |
文本识别模型
中文识别模型
| 模型名称 | 模型简介 | 配置文件 | 推理模型大小 | 下载地址 |
|---|---|---|---|---|
| ch_ppocr_mobile_slim_v2.0_rec | slim裁剪量化版超轻量模型，支持中英文、数字识别 | rec_chinese_lite_train_v2.0.yml | 6M | 推理模型 / 训练模型 |
| ch_ppocr_mobile_v2.0_rec | 原始超轻量模型，支持中英文、数字识别 | rec_chinese_lite_train_v2.0.yml | 5.2M | 推理模型 / 训练模型 / 预训练模型 |
| ch_ppocr_server_v2.0_rec | 通用模型，支持中英文、数字识别 | rec_chinese_common_train_v2.0.yml | 94.8M | 推理模型 / 训练模型 / 预训练模型 |
说明： 训练模型是基于预训练模型在真实数据与竖排合成文本数据上finetune得到的模型，在真实应用场景中有着更好的表现，预训练模型则是直接基于全量真实数据与合成数据训练得到，更适合用于在自己的数据集上finetune。
英文识别模型
| 模型名称 | 模型简介 | 配置文件 | 推理模型大小 | 下载地址 |
|---|---|---|---|---|
| en_number_mobile_slim_v2.0_rec | slim裁剪量化版超轻量模型，支持英文、数字识别 | rec_en_number_lite_train.yml | 2.7M | 推理模型 / 训练模型 |
| en_number_mobile_v2.0_rec | 原始超轻量模型，支持英文、数字识别 | rec_en_number_lite_train.yml | 2.6M | 推理模型 / 训练模型 |
%cd ~/PaddleOCR/
# mobile模型
# !wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar
# !tar -xf ch_ppocr_mobile_v2.0_rec_pre.tar
# server模型
!wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar
!tar -xf ch_ppocr_server_v2.0_rec_pre.tar
/home/aistudio/PaddleOCR
--2021-12-31 12:58:03-- https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar
Resolving paddleocr.bj.bcebos.com (paddleocr.bj.bcebos.com)... 182.61.200.195, 182.61.200.229, 2409:8c04:1001:1002:0:ff:b001:368a
Connecting to paddleocr.bj.bcebos.com (paddleocr.bj.bcebos.com)|182.61.200.195|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 490184704 (467M) [application/x-tar]
Saving to: ‘ch_ppocr_server_v2.0_rec_pre.tar’
ch_ppocr_server_v2. 100%[===================>] 467.48M 56.0MB/s in 13s
2021-12-31 12:58:17 (34.9 MB/s) - ‘ch_ppocr_server_v2.0_rec_pre.tar’ saved [490184704/490184704]
五、训练
%cd ~/PaddleOCR/
# mobile模型
# !python tools/train.py -c ./configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.checkpoints=./output/rec_chinese_lite_v2.0/latest
# server模型
!python tools/train.py -c ./configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml
1.选择合适的batch size
2.训练日志
[2021/12/30 23:26:54] root INFO: epoch: [68/500], iter: 9930, lr: 0.000962, loss: 5.635038, acc: 0.521482, norm_edit_dis: 0.745346, reader_cost: 0.01405 s, batch_cost: 0.26990 s, samples: 2560, ips: 948.50786
[2021/12/30 23:27:11] root INFO: epoch: [68/500], iter: 9940, lr: 0.000962, loss: 5.653114, acc: 0.509764, norm_edit_dis: 0.740487, reader_cost: 0.01402 s, batch_cost: 0.26862 s, samples: 2560, ips: 953.03473
[2021/12/30 23:27:26] root INFO: epoch: [68/500], iter: 9950, lr: 0.000962, loss: 5.411234, acc: 0.515623, norm_edit_dis: 0.748549, reader_cost: 0.00091 s, batch_cost: 0.26371 s, samples: 2560, ips: 970.76457
[2021/12/30 23:27:40] root INFO: epoch: [68/500], iter: 9960, lr: 0.000962, loss: 5.588465, acc: 0.525389, norm_edit_dis: 0.755345, reader_cost: 0.00684 s, batch_cost: 0.25901 s, samples: 2560, ips: 988.38445
[2021/12/30 23:27:48] root INFO: epoch: [68/500], iter: 9970, lr: 0.000961, loss: 5.789876, acc: 0.513670, norm_edit_dis: 0.740609, reader_cost: 0.00095 s, batch_cost: 0.15022 s, samples: 2560, ips: 1704.17763
[2021/12/30 23:27:51] root INFO: epoch: [68/500], iter: 9974, lr: 0.000961, loss: 5.787237, acc: 0.511717, norm_edit_dis: 0.747102, reader_cost: 0.00018 s, batch_cost: 0.05935 s, samples: 1024, ips: 1725.41448
[2021/12/30 23:27:51] root INFO: save model in ./output/rec_chinese_lite_v2.0/latest
[2021/12/30 23:27:51] root INFO: Initialize indexs of datasets:['/home/aistudio/data/data62842/train.txt']
[2021/12/30 23:28:21] root INFO: epoch: [69/500], iter: 9980, lr: 0.000961, loss: 5.801509, acc: 0.517576, norm_edit_dis: 0.749756, reader_cost: 1.10431 s, batch_cost: 1.37585 s, samples: 1536, ips: 111.64048
[2021/12/30 23:28:40] root INFO: epoch: [69/500], iter: 9990, lr: 0.000961, loss: 5.548770, acc: 0.533201, norm_edit_dis: 0.762078, reader_cost: 0.00839 s, batch_cost: 0.32035 s, samples: 2560, ips: 799.11578
[2021/12/30 23:28:56] root INFO: epoch: [69/500], iter: 10000, lr: 0.000961, loss: 5.449094, acc: 0.537107, norm_edit_dis: 0.762517, reader_cost: 0.00507 s, batch_cost: 0.25845 s, samples: 2560, ips: 990.51517
eval model:: 100%|██████████████████████████████| 20/20 [00:15<00:00, 1.98it/s]
[2021/12/30 23:29:12] root INFO: cur metric, acc: 0.4641999071600186, norm_edit_dis: 0.6980459628854201, fps: 4204.853978632389
[2021/12/30 23:29:12] root INFO: best metric, acc: 0.48179990364001923, start_epoch: 12, norm_edit_dis: 0.7096561279006699, fps: 4618.199275059127, best_epoch: 46
3. visualdl可视化
- 本地安装visualdl
pip install visualdl - 下载日志至本地
- 启动visualdl可视化
visualdl --logdir ./ - 打开浏览器查看
http://localhost:8040/
六、模型评估
# GPU 评估， Global.checkpoints 为待测权重
%cd ~/PaddleOCR/
# mobile模型
# !python -m paddle.distributed.launch tools/eval.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml \
# -o Global.checkpoints=./output/rec_chinese_lite_v2.0/latest
# server模型
!python -m paddle.distributed.launch tools/eval.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml \
-o Global.checkpoints=./output/rec_chinese_common_v2.0/best_accuracy.pdparams
/home/aistudio/PaddleOCR
----------- Configuration Arguments -----------
backend: auto
elastic_server: None
force: False
gpus: None
heter_devices:
heter_worker_num: None
heter_workers:
host: None
http_port: None
ips: 127.0.0.1
job_id: None
log_dir: log
np: None
nproc_per_node: None
run_mode: None
scale: 0
server_num: None
servers:
training_script: tools/eval.py
training_script_args: ['-c', 'configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml', '-o', 'Global.checkpoints=./output/rec_chinese_common_v2.0/best_accuracy.pdparams']
worker_num: None
workers:
------------------------------------------------
WARNING 2021-12-31 18:51:19,722 launch.py:423] Not found distinct arguments and compiled with cuda or xpu. Default use collective mode
launch train in GPU mode!
INFO 2021-12-31 18:51:19,725 launch_utils.py:528] Local start 1 processes. First process distributed environment info (Only For Debug):
+=======================================================================================+
| Distributed Envs Value |
+---------------------------------------------------------------------------------------+
| PADDLE_TRAINER_ID 0 |
| PADDLE_CURRENT_ENDPOINT 127.0.0.1:46063 |
| PADDLE_TRAINERS_NUM 1 |
| PADDLE_TRAINER_ENDPOINTS 127.0.0.1:46063 |
| PADDLE_RANK_IN_NODE 0 |
| PADDLE_LOCAL_DEVICE_IDS 0 |
| PADDLE_WORLD_DEVICE_IDS 0 |
| FLAGS_selected_gpus 0 |
| FLAGS_selected_accelerators 0 |
+=======================================================================================+
INFO 2021-12-31 18:51:19,725 launch_utils.py:532] details abouts PADDLE_TRAINER_ENDPOINTS can be found in log/endpoints.log, and detail running logs maybe found in log/workerlog.0
launch proc_id:16263 idx:0
[2021/12/31 18:51:21] root INFO: Architecture :
[2021/12/31 18:51:21] root INFO: Backbone :
[2021/12/31 18:51:21] root INFO: layers : 34
[2021/12/31 18:51:21] root INFO: name : ResNet
[2021/12/31 18:51:21] root INFO: Head :
[2021/12/31 18:51:21] root INFO: fc_decay : 4e-05
[2021/12/31 18:51:21] root INFO: name : CTCHead
[2021/12/31 18:51:21] root INFO: Neck :
[2021/12/31 18:51:21] root INFO: encoder_type : rnn
[2021/12/31 18:51:21] root INFO: hidden_size : 256
[2021/12/31 18:51:21] root INFO: name : SequenceEncoder
[2021/12/31 18:51:21] root INFO: Transform : None
[2021/12/31 18:51:21] root INFO: algorithm : CRNN
[2021/12/31 18:51:21] root INFO: model_type : rec
[2021/12/31 18:51:21] root INFO: Eval :
[2021/12/31 18:51:21] root INFO: dataset :
[2021/12/31 18:51:21] root INFO: data_dir : /home/aistudio/data/data62842/train_images
[2021/12/31 18:51:21] root INFO: label_file_list : ['/home/aistudio/data/data62842/val.txt']
[2021/12/31 18:51:21] root INFO: name : SimpleDataSet
[2021/12/31 18:51:21] root INFO: transforms :
[2021/12/31 18:51:21] root INFO: DecodeImage :
[2021/12/31 18:51:21] root INFO: channel_first : False
[2021/12/31 18:51:21] root INFO: img_mode : BGR
[2021/12/31 18:51:21] root INFO: CTCLabelEncode : None
[2021/12/31 18:51:21] root INFO: RecResizeImg :
[2021/12/31 18:51:21] root INFO: image_shape : [3, 32, 320]
[2021/12/31 18:51:21] root INFO: KeepKeys :
[2021/12/31 18:51:21] root INFO: keep_keys : ['image', 'label', 'length']
[2021/12/31 18:51:21] root INFO: loader :
[2021/12/31 18:51:21] root INFO: batch_size_per_card : 256
[2021/12/31 18:51:21] root INFO: drop_last : False
[2021/12/31 18:51:21] root INFO: num_workers : 8
[2021/12/31 18:51:21] root INFO: shuffle : False
[2021/12/31 18:51:21] root INFO: Global :
[2021/12/31 18:51:21] root INFO: cal_metric_during_train : True
[2021/12/31 18:51:21] root INFO: character_dict_path : ppocr/utils/ppocr_keys_v1.txt
[2021/12/31 18:51:21] root INFO: checkpoints : ./output/rec_chinese_common_v2.0/best_accuracy.pdparams
[2021/12/31 18:51:21] root INFO: debug : False
[2021/12/31 18:51:21] root INFO: distributed : False
[2021/12/31 18:51:21] root INFO: epoch_num : 500
[2021/12/31 18:51:21] root INFO: eval_batch_step : [0, 2000]
[2021/12/31 18:51:21] root INFO: infer_img : doc/imgs_words/ch/word_1.jpg
[2021/12/31 18:51:21] root INFO: infer_mode : False
[2021/12/31 18:51:21] root INFO: log_smooth_window : 20
[2021/12/31 18:51:21] root INFO: max_text_length : 25
[2021/12/31 18:51:21] root INFO: pretrained_model : ./ch_ppocr_server_v2.0_rec_pre/best_accuracy
[2021/12/31 18:51:21] root INFO: print_batch_step : 10
[2021/12/31 18:51:21] root INFO: save_epoch_step : 3
[2021/12/31 18:51:21] root INFO: save_inference_dir : None
[2021/12/31 18:51:21] root INFO: save_model_dir : ./output/rec_chinese_common_v2.0
[2021/12/31 18:51:21] root INFO: save_res_path : ./output/rec/predicts_chinese_common_v2.0.txt
[2021/12/31 18:51:21] root INFO: use_gpu : True
[2021/12/31 18:51:21] root INFO: use_space_char : True
[2021/12/31 18:51:21] root INFO: use_visualdl : False
[2021/12/31 18:51:21] root INFO: Loss :
[2021/12/31 18:51:21] root INFO: name : CTCLoss
[2021/12/31 18:51:21] root INFO: Metric :
[2021/12/31 18:51:21] root INFO: main_indicator : acc
[2021/12/31 18:51:21] root INFO: name : RecMetric
[2021/12/31 18:51:21] root INFO: Optimizer :
[2021/12/31 18:51:21] root INFO: beta1 : 0.9
[2021/12/31 18:51:21] root INFO: beta2 : 0.999
[2021/12/31 18:51:21] root INFO: lr :
[2021/12/31 18:51:21] root INFO: learning_rate : 0.001
[2021/12/31 18:51:21] root INFO: name : Cosine
[2021/12/31 18:51:21] root INFO: warmup_epoch : 5
[2021/12/31 18:51:21] root INFO: name : Adam
[2021/12/31 18:51:21] root INFO: regularizer :
[2021/12/31 18:51:21] root INFO: factor : 4e-05
[2021/12/31 18:51:21] root INFO: name : L2
[2021/12/31 18:51:21] root INFO: PostProcess :
[2021/12/31 18:51:21] root INFO: name : CTCLabelDecode
[2021/12/31 18:51:21] root INFO: Train :
[2021/12/31 18:51:21] root INFO: dataset :
[2021/12/31 18:51:21] root INFO: data_dir : /home/aistudio/data/data62842/train_images
[2021/12/31 18:51:21] root INFO: label_file_list : ['/home/aistudio/data/data62842/train.txt']
[2021/12/31 18:51:21] root INFO: name : SimpleDataSet
[2021/12/31 18:51:21] root INFO: transforms :
[2021/12/31 18:51:21] root INFO: DecodeImage :
[2021/12/31 18:51:21] root INFO: channel_first : False
[2021/12/31 18:51:21] root INFO: img_mode : BGR
[2021/12/31 18:51:21] root INFO: RecAug : None
[2021/12/31 18:51:21] root INFO: CTCLabelEncode : None
[2021/12/31 18:51:21] root INFO: RecResizeImg :
[2021/12/31 18:51:21] root INFO: image_shape : [3, 32, 320]
[2021/12/31 18:51:21] root INFO: KeepKeys :
[2021/12/31 18:51:21] root INFO: keep_keys : ['image', 'label', 'length']
[2021/12/31 18:51:21] root INFO: loader :
[2021/12/31 18:51:21] root INFO: batch_size_per_card : 256
[2021/12/31 18:51:21] root INFO: drop_last : True
[2021/12/31 18:51:21] root INFO: num_workers : 8
[2021/12/31 18:51:21] root INFO: shuffle : True
[2021/12/31 18:51:21] root INFO: profiler_options : None
[2021/12/31 18:51:21] root INFO: train with paddle 2.2.1 and device CUDAPlace(0)
[2021/12/31 18:51:21] root INFO: Initialize indexs of datasets:['/home/aistudio/data/data62842/val.txt']
W1231 18:51:21.482865 16263 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 10.1, Runtime API Version: 10.1
W1231 18:51:21.487445 16263 device_context.cc:465] device: 0, cuDNN Version: 7.6.
[2021/12/31 18:51:26] root INFO: resume from ./output/rec_chinese_common_v2.0/best_accuracy
[2021/12/31 18:51:26] root INFO: metric in ckpt ***************
[2021/12/31 18:51:26] root INFO: acc:0.6035998792800241
[2021/12/31 18:51:26] root INFO: norm_edit_dis:0.8053270782756357
[2021/12/31 18:51:26] root INFO: fps:438.9587163608945
[2021/12/31 18:51:26] root INFO: best_epoch:23
[2021/12/31 18:51:26] root INFO: start_epoch:24
eval model:: 0%| | 0/20 [00:00<?, ?it/s]
eval model:: 5%|▌ | 1/20 [00:02<00:54, 2.88s/it]
eval model:: 10%|█ | 2/20 [00:04<00:43, 2.42s/it]
eval model:: 15%|█▌ | 3/20 [00:05<00:35, 2.10s/it]
eval model:: 20%|██ | 4/20 [00:06<00:29, 1.87s/it]
eval model:: 25%|██▌ | 5/20 [00:08<00:25, 1.71s/it]
eval model:: 30%|███ | 6/20 [00:09<00:22, 1.60s/it]
eval model:: 35%|███▌ | 7/20 [00:10<00:19, 1.53s/it]
eval model:: 40%|████ | 8/20 [00:12<00:17, 1.48s/it]
eval model:: 45%|████▌ | 9/20 [00:13<00:15, 1.44s/it]
eval model:: 50%|█████ | 10/20 [00:15<00:14, 1.42s/it]
eval model:: 55%|█████▌ | 11/20 [00:16<00:12, 1.40s/it]
eval model:: 60%|██████ | 12/20 [00:17<00:11, 1.39s/it]
eval model:: 65%|██████▌ | 13/20 [00:19<00:09, 1.38s/it]
eval model:: 70%|███████ | 14/20 [00:20<00:08, 1.38s/it]
eval model:: 75%|███████▌ | 15/20 [00:21<00:06, 1.38s/it]
eval model:: 80%|████████ | 16/20 [00:23<00:05, 1.38s/it]
eval model:: 85%|████████▌ | 17/20 [00:24<00:04, 1.38s/it]
eval model:: 90%|█████████ | 18/20 [00:25<00:02, 1.37s/it]
eval model:: 95%|█████████▌| 19/20 [00:27<00:01, 1.37s/it]
eval model:: 100%|██████████| 20/20 [00:28<00:00, 1.17s/it]
[2021/12/31 18:51:54] root INFO: metric eval ***************
[2021/12/31 18:51:54] root INFO: acc:0.6035998792800241
[2021/12/31 18:51:54] root INFO: norm_edit_dis:0.8053270782756357
[2021/12/31 18:51:54] root INFO: fps:439.3796693669832
INFO 2021-12-31 18:51:55,788 launch.py:311] Local processes completed.
七、结果预测
预测脚本使用预测训练好的模型，并将结果保存成txt格式，可以直接送到比赛提交入口测评，文件默认保存在output/rec/predicts_chinese_lite_v2.0.txt
1.提交内容与格式
本次比赛要求参赛选手必须提交使用深度学习平台飞桨（PaddlePaddle）训练的模型。参赛者要求以.txt 文本格式提交结果，其中每一行是图片名称和文字预测的结果，中间以 “\t” 作为分割符，示例如下：
| new_name | value |
|---|---|
| 0.jpg | 文本0 |
2. infer_rec.py修改
with open(save_res_path, "w") as fout:
#添加列头
fout.write('new_name' + "\t" + 'value' +'\n')
for file in get_image_file_list(config['Global']['infer_img']):
logger.info("infer_img: {}".format(file))
with open(file, 'rb') as f:
img = f.read()
data = {'image': img}
batch = transform(data, ops)
if config['Architecture']['algorithm'] == "SRN":
encoder_word_pos_list = np.expand_dims(batch[1], axis=0)
gsrm_word_pos_list = np.expand_dims(batch[2], axis=0)
gsrm_slf_attn_bias1_list = np.expand_dims(batch[3], axis=0)
gsrm_slf_attn_bias2_list = np.expand_dims(batch[4], axis=0)
others = [
paddle.to_tensor(encoder_word_pos_list),
paddle.to_tensor(gsrm_word_pos_list),
paddle.to_tensor(gsrm_slf_attn_bias1_list),
paddle.to_tensor(gsrm_slf_attn_bias2_list)
]
if config['Architecture']['algorithm'] == "SAR":
valid_ratio = np.expand_dims(batch[-1], axis=0)
img_metas = [paddle.to_tensor(valid_ratio)]
images = np.expand_dims(batch[0], axis=0)
images = paddle.to_tensor(images)
if config['Architecture']['algorithm'] == "SRN":
preds = model(images, others)
elif config['Architecture']['algorithm'] == "SAR":
preds = model(images, img_metas)
else:
preds = model(images)
post_result = post_process_class(preds)
info = None
if isinstance(post_result, dict):
rec_info = dict()
for key in post_result:
if len(post_result[key][0]) >= 2:
rec_info[key] = {
"label": post_result[key][0][0],
"score": float(post_result[key][0][1]),
}
info = json.dumps(rec_info)
else:
if len(post_result[0]) >= 2:
info = post_result[0][0] + "\t" + str(post_result[0][1])
if info is not None:
logger.info("\t result: {}".format(info))
# fout.write(file + "\t" + info)
# 格式化输出
fout.write(file + "\t" + post_result[0][0] +'\n')
logger.info("success!")
%cd ~/PaddleOCR/
# mobile模型
# !python tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml \
# -o Global.infer_img="/home/aistudio/data/data62843/test_images" \
# Global.pretrained_model="./output/rec_chinese_lite_v2.0/best_accuracy"
# server模型
!python tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml \
-o Global.infer_img="/home/aistudio/data/data62843/test_images" \
Global.checkpoints=./output/rec_chinese_common_v2.0/best_accuracy
预测日志
[2021/12/30 23:53:50] root INFO: result: 萧记果点 0.66611135
[2021/12/30 23:53:50] root INFO: infer_img: /home/aistudio/data/data62843/test_images/9995.jpg
[2021/12/30 23:53:50] root INFO: result: 福 0.1693737
[2021/12/30 23:53:50] root INFO: infer_img: /home/aistudio/data/data62843/test_images/9996.jpg
[2021/12/30 23:53:50] root INFO: result: 279 0.97771764
[2021/12/30 23:53:50] root INFO: infer_img: /home/aistudio/data/data62843/test_images/9997.jpg
[2021/12/30 23:53:50] root INFO: result: 公牛装饰开关 0.9916236
[2021/12/30 23:53:50] root INFO: infer_img: /home/aistudio/data/data62843/test_images/9998.jpg
[2021/12/30 23:53:50] root INFO: result: 专酒 0.118371546
[2021/12/30 23:53:50] root INFO: infer_img: /home/aistudio/data/data62843/test_images/9999.jpg
[2021/12/30 23:53:50] root INFO: result: 东之家 0.871051
[2021/12/30 23:53:50] root INFO: success!
...
...
八、基于预测引擎的预测
1.模型大小限制
约束性条件1：模型总大小不超过10MB（以.pdmodel和.pdiparams文件非压缩状态磁盘占用空间之和为准）；
2.解决办法
训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。实际上，此处的约束条件限制的是inference 模型的大小。inference 模型一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成，模型大小也会小一些。
# 静态模型导出
%cd ~/PaddleOCR/
# mobile模型
# !python tools/export_model.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model=./output/rec_chinese_lite_v2.0/best_accuracy.pdparams Global.save_inference_dir=./inference/rec_inference/
# server模型
!python tools/export_model.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_common_v2.0.yml -o Global.pretrained_model=./output/rec_chinese_common_train_v2.0/best_accuracy.pdparams Global.save_inference_dir=./inference/rec_inference/
/home/aistudio/PaddleOCR
W1230 23:54:48.747483 13346 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
W1230 23:54:48.752360 13346 device_context.cc:465] device: 0, cuDNN Version: 7.6.
[2021/12/30 23:54:52] root INFO: load pretrain successful from ./output/rec_chinese_lite_v2.0/best_accuracy
[2021/12/30 23:54:54] root INFO: inference model is saved to ./inference/rec_inference/inference
%cd ~/PaddleOCR/
!du -sh ./inference/rec_inference/
/home/aistudio/PaddleOCR
5.2M ./inference/rec_inference/
- 可以看到，当前训练使用的CRNN算法导出inference后，仅有5.2M。
- 导出的inference模型也可以用来预测，预测逻辑如下代码所示。
# 使用导出静态模型预测
%cd ~/PaddleOCR/
!python3.7 tools/infer/predict_rec.py --rec_model_dir=./inference/rec_inference/ --image_dir="/home/aistudio/data/A榜测试数据集/TestAImages"
预测日志
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000001.jpg:('MJ', 0.2357887)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000002.jpg:('中门', 0.7167614)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000003.jpg:('黄焖鸡米饭', 0.7325407)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000004.jpg:('加行', 0.06699998)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000005.jpg:('学商烤面航', 0.40579563)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000006.jpg:('绿村装机 滋光彩机 CP口出国', 0.38243735)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000007.jpg:('有酸锁 四好吃', 0.38957664)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000008.jpg:('婚汽中海', 0.36037388)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000009.jpg:('L', 0.25453746)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000010.jpg:('清女装', 0.79736567)
[2021/12/30 13:20:08] root INFO: Predicts of /home/aistudio/data/A榜测试数据集/TestAImages/TestA_000011.jpg:('幼小数学视食', 0.50577885)
...
...
九、提交
预测结果保存到配置文件指定的 output/rec/predicts_chinese_lite_v2.0.txt文件，可直接提交即可。
%cd ~
!head PaddleOCR/output/rec/predicts_chinese_common_v2.0.txt
/home/aistudio
new_name value
0.jpg 邦佳洗衣
1.jpg 不锈钢配件大全
10.jpg 诊疗科目：中医科
100.jpg 210
1000.jpg 电线电缆等
1001.jpg 20
1002.jpg 进口滤纸 专业制造
1003.jpg 1506540
1004.jpg ｉＷｏＷ
210
1000.jpg 电线电缆等
1001.jpg 20
1002.jpg 进口滤纸 专业制造
1003.jpg 1506540
1004.jpg ｉＷｏＷ
1.mobile模型
2.server模型
大家可以再处理处理，优化优化，多跑几轮。
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
0
0- 0
已为社区贡献1438条内容
所有评论(0)