Paddle打比赛】基于PaddleOCR的AIWIN手写体识别
使用PaddleOCR的 CRNN+CTC对大写金额数字、日期数字进行识别。
·
一、【Paddle打比赛】AIWIN 手写体OCR识别竞赛
1.赛题背景
http://ailab.aiwin.org.cn/competitions/65#learn_the_details
银行日常业务中涉及到各类凭证的识别录入，例如身份证录入、支票录入、对账单录入等。以往的录入方式主要是以人工录入为主，效率较低，人力成本较高。近几年来，OCR相关技术以其自动执行、人为干预较少等特点正逐步替代传统的人工录入方式。但OCR技术在实际应用中也存在一些问题，在各类凭证字段的识别中，手写体由于其字体差异性大、字数不固定、语义关联性较低、凭证背景干扰等原因，导致OCR识别率准确率不高，需要大量人工校正，对日常的银行录入业务造成了一定的影响。
2. 赛题任务
本次赛题将提供手写体图像切片数据集，数据集从真实业务场景中，经过切片脱敏得到，参赛队伍通过识别技术，获得对应的识别结果。即：
输入：手写体图像切片数据集
输出：对应的识别结果
赛题在赛程中分设为两个独立任务，各自设定不同条件的训练集、测试集和建模环境，概述如下：
- 任务一：提供开放可下载的训练集及测试集，允许线下建模或线上提供 Notebook 环境及 Terminal 容器环境（脱网）建模，输出识别结果完成赛题。
- 任务二：提供不可下载的训练集，要求线上通过 Terminal 容器环境（脱网）建模后提交模型，由系统输入测试集（即对选手不可见），输出识别结果完成赛题。
3.数据基本情况
|
| 任务一 | 任务二 |
| 训练集（含验证集，请自行划分） | 8 千张图像，包含年份、金额2种信息 | 3 万张图像，包含银行名称、年份、月份、日期、金额5 种信息。 |
| 测试集 | 2 千张图像 | 设定 AB榜： A 榜：5 千张图像 B 榜：5 千张图像 |
原始手写体图像共分为三类，分别涉及银行名称、年月日、金额三大类，分别示意如下：
相应图片切片中可能混杂有一定量的干扰信息，分别示例如下；
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
三、数据准备
主要任务有：
- 数据解压缩
- det数据集格式化和数据集划分
- rec数据集格式化和数据集划分
# 解压缩
%cd ~
!unzip -qoa 2021A_T1_Task1_数据集含训练集和测试集.zip
/home/aistudio
# 查看数据集
from PIL import Image
img=Image.open("训练集/amount/images/8bb39447774eb21a01777a9efa890543.jpg")
img
![[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-FODMTaYp-1642075460147)(output_6_0.png)]](https://img-blog.csdnimg.cn/e46601c1a3d54438972e09fed898c5b7.png)
1.数额数据处理
%cd ~
# 查看数据集
!head 训练集/amount/gt.json
/home/aistudio
{
"8bb39426774ee53f017770203bab0bc5.jpg": "肆佰肆拾贰元整",
"8bb39447760a31c801762283f9dd63cb.jpg": "贰仟壹佰壹拾贰元整",
"8bb1943d774eb211017784b7af783c23.jpg": "壹仟肆佰贰拾元整",
"8bb194277657bb0501768d5379a4262b.jpg": "伍佰叁拾捌元叁角捌分",
"8bb3942b7657bb83017674d349786868.jpg": "肆佰元整",
"8bb1943d760a31b70176275a31832557.jpg": "壹万贰仟贰佰元整",
"8bb19437760a2b5f017641f9743b41b4.jpg": "叁万捌仟肆佰伍拾捌元捌角捌分",
"8bb1941c7657bb01017674b446cc2a2e.jpg": "贰仟肆佰捌拾陆元整",
"8bb39441760a31b601764a13149e3008.jpg": "玖仟肆佰伍拾元整",
import glob, codecs, json, os
import numpy as np
amount_jpgs = glob.glob('./训练集/amount/images/*.jpg')
lines = codecs.open('./训练集/amount/gt.json', encoding='utf-8').readlines()
lines = ''.join(lines)
amount_gt = json.loads(lines.replace(',\n}', '}'))
# 划分train和eval
%cd ~/
# 写入列表文件
f_train=open("./训练集/amount/train_list.txt", 'w')
f_val=open("./训练集/amount/val_list.txt", 'w')
i=0
for key in amount_gt:
if i%10==0:
f_val.write(key+ '\t'+amount_gt[key]+'\n')
else:
f_train.write(key+ '\t'+amount_gt[key]+'\n')
i=i+1
/home/aistudio
!head ./训练集/amount/train_list.txt
8bb39447760a31c801762283f9dd63cb.jpg 贰仟壹佰壹拾贰元整
8bb1943d774eb211017784b7af783c23.jpg 壹仟肆佰贰拾元整
8bb194277657bb0501768d5379a4262b.jpg 伍佰叁拾捌元叁角捌分
8bb3942b7657bb83017674d349786868.jpg 肆佰元整
8bb1943d760a31b70176275a31832557.jpg 壹万贰仟贰佰元整
8bb19437760a2b5f017641f9743b41b4.jpg 叁万捌仟肆佰伍拾捌元捌角捌分
8bb1941c7657bb01017674b446cc2a2e.jpg 贰仟肆佰捌拾陆元整
8bb39441760a31b601764a13149e3008.jpg 玖仟肆佰伍拾元整
8bb3943c774eb20601775cf697f0456b.jpg 壹万柒仟贰佰叁拾壹元整
8bb194207657bb1201765fd7645934b5.jpg 肆仟伍佰壹拾叁元贰角整
# 生成字典
amount_gt.update(amount_gt)
s = ''
for x in amount_gt:
s += amount_gt[x]
char_list = list(set(list(s)))
char_list = char_list
with open('./训练集/amount/vocabulary.txt', 'w') as up:
for x in char_list:
up.write(x + '\n')
!cat ./训练集/amount/vocabulary.txt
分
亿
圆
零
万
元
陆
角
柒
贰
捌
玖
拾
佰
整
正
壹
肆
伍
仟
叁
2.日期数据处理
%cd ~
date_jpgs = glob.glob('./训练集/date/images/*.jpg')
lines = codecs.open('./训练集/date/gt.json', encoding='utf-8').readlines()
lines = ''.join(lines)
date_gt = json.loads(lines.replace(',\n}', '}'))
/home/aistudio
# 划分train和eval
# 写入列表文件
f_train=open("./训练集/date/train_list.txt", 'w')
f_val=open("./训练集/date/val_list.txt", 'w')
i=0
for key in date_gt:
if i%10==0:
f_val.write(key+ '\t'+date_gt[key]+'\n')
else:
f_train.write(key+ '\t'+date_gt[key]+'\n')
i=i+1
date_gt.update(date_gt)
s = ''
for x in date_gt:
s += date_gt[x]
char_list = list(set(list(s)))
char_list = char_list
with open('./训练集/date/vocabulary.txt', 'w') as up:
for x in char_list:
up.write(x + '\n')
!cat ./训练集/date/vocabulary.txt
陆
贰
捌
壹
玖
肆
伍
零
柒
拾
叁
四、金额训练与评估
以PaddleOCR/configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml为基准进行配置
1. 金额训练
Global:
use_gpu: true
epoch_num: 500
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/rec_chinese_common_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:
checkpoints:
save_inference_dir:
use_visualdl: False
infer_img: doc/imgs_words/ch/word_1.jpg
# for data or label process
character_dict_path: /home/aistudio/训练集/amount/vocabulary.txt
max_text_length: 25
infer_mode: False
use_space_char: True
save_res_path: ./output/rec/predicts_chinese_common_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: 0.00004
Architecture:
model_type: rec
algorithm: CRNN
Transform:
Backbone:
name: ResNet
layers: 34
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 256
Head:
name: CTCHead
fc_decay: 0.00004
Loss:
name: CTCLoss
PostProcess:
name: CTCLabelDecode
Metric:
name: RecMetric
main_indicator: acc
Train:
dataset:
name: SimpleDataSet
data_dir: /home/aistudio/训练集/amount/images
label_file_list: ["/home/aistudio/训练集/amount/train_list.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/训练集/amount/images
label_file_list: ["/home/aistudio/训练集/amount/val_list.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
2.下载预训练模型
%cd ~/PaddleOCR/
# 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
--2022-01-09 19:54:27-- 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)... 182.61.200.229, 182.61.200.195, 2409:8c04:1001:1002:0:ff:b001:368a
Connecting to paddleocr.bj.bcebos.com (paddleocr.bj.bcebos.com)|182.61.200.229|: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 48.1MB/s in 13s
2022-01-09 19:54:40 (35.6 MB/s) - ‘ch_ppocr_server_v2.0_rec_pre.tar’ saved [490184704/490184704]
3.金额训练
# 覆盖配置文件
!cp -f ../rec_chinese_common_train_v2.0.yml ./configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml
# server模型
%cd ~/PaddleOCR/
!python tools/train.py -c ./configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml
训练日志
[2022/01/09 20:21:27] root INFO: epoch: [35/500], iter: 480, lr: 0.000992, loss: 0.322123, acc: 0.972652, norm_edit_dis: 0.994506, reader_cost: 0.57496 s, batch_cost: 0.89715 s, samples: 1280, ips: 142.67360
[2022/01/09 20:21:43] root INFO: epoch: [35/500], iter: 489, lr: 0.000992, loss: 0.256427, acc: 0.974606, norm_edit_dis: 0.995029, reader_cost: 0.00015 s, batch_cost: 0.55269 s, samples: 2304, ips: 416.86967
[2022/01/09 20:21:47] root INFO: save model in ./output/rec_chinese_common_v2.0/latest
[2022/01/09 20:21:47] root INFO: Initialize indexs of datasets:['/home/aistudio/训练集/amount/train_list.txt']
[2022/01/09 20:21:55] root INFO: epoch: [36/500], iter: 490, lr: 0.000992, loss: 0.249966, acc: 0.974606, norm_edit_dis: 0.995029, reader_cost: 0.63610 s, batch_cost: 0.69958 s, samples: 256, ips: 36.59348
[2022/01/09 20:22:13] root INFO: epoch: [36/500], iter: 500, lr: 0.000991, loss: 0.246262, acc: 0.968746, norm_edit_dis: 0.995142, reader_cost: 0.00045 s, batch_cost: 0.61365 s, samples: 2560, ips: 417.17631
eval model:: 100%|████████████████████████████████| 2/2 [00:02<00:00, 1.04s/it]
[2022/01/09 20:22:15] root INFO: cur metric, acc: 0.9699975750060625, norm_edit_dis: 0.9909632884925447, fps: 456.3268190347603
4.金额模型评估
# 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
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
def convert_to_list(value, n, name, dtype=np.int):
----------- Configuration Arguments -----------
gpus: None
heter_worker_num: None
heter_workers:
http_port: None
ips: 127.0.0.1
log_dir: log
nproc_per_node: None
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 2022-01-09 20:25:44,789 launch.py:316] Not found distinct arguments and compiled with cuda. Default use collective mode
launch train in GPU mode
INFO 2022-01-09 20:25:44,793 launch_utils.py:471] Local start 1 processes. First process distributed environment info (Only For Debug):
+=======================================================================================+
| Distributed Envs Value |
+---------------------------------------------------------------------------------------+
| PADDLE_TRAINER_ID 0 |
| PADDLE_CURRENT_ENDPOINT 127.0.0.1:45561 |
| PADDLE_TRAINERS_NUM 1 |
| PADDLE_TRAINER_ENDPOINTS 127.0.0.1:45561 |
| FLAGS_selected_gpus 0 |
+=======================================================================================+
INFO 2022-01-09 20:25:44,793 launch_utils.py:475] details abouts PADDLE_TRAINER_ENDPOINTS can be found in log/endpoints.log, and detail running logs maybe found in log/workerlog.0
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
def convert_to_list(value, n, name, dtype=np.int):
[2022/01/09 20:25:46] root INFO: Architecture :
[2022/01/09 20:25:46] root INFO: Backbone :
[2022/01/09 20:25:46] root INFO: layers : 34
[2022/01/09 20:25:46] root INFO: name : ResNet
[2022/01/09 20:25:46] root INFO: Head :
[2022/01/09 20:25:46] root INFO: fc_decay : 4e-05
[2022/01/09 20:25:46] root INFO: name : CTCHead
[2022/01/09 20:25:46] root INFO: Neck :
[2022/01/09 20:25:46] root INFO: encoder_type : rnn
[2022/01/09 20:25:46] root INFO: hidden_size : 256
[2022/01/09 20:25:46] root INFO: name : SequenceEncoder
[2022/01/09 20:25:46] root INFO: Transform : None
[2022/01/09 20:25:46] root INFO: algorithm : CRNN
[2022/01/09 20:25:46] root INFO: model_type : rec
[2022/01/09 20:25:46] root INFO: Eval :
[2022/01/09 20:25:46] root INFO: dataset :
[2022/01/09 20:25:46] root INFO: data_dir : /home/aistudio/训练集/amount/images
[2022/01/09 20:25:46] root INFO: label_file_list : ['/home/aistudio/训练集/amount/val_list.txt']
[2022/01/09 20:25:46] root INFO: name : SimpleDataSet
[2022/01/09 20:25:46] root INFO: transforms :
[2022/01/09 20:25:46] root INFO: DecodeImage :
[2022/01/09 20:25:46] root INFO: channel_first : False
[2022/01/09 20:25:46] root INFO: img_mode : BGR
[2022/01/09 20:25:46] root INFO: CTCLabelEncode : None
[2022/01/09 20:25:46] root INFO: RecResizeImg :
[2022/01/09 20:25:46] root INFO: image_shape : [3, 32, 320]
[2022/01/09 20:25:46] root INFO: KeepKeys :
[2022/01/09 20:25:46] root INFO: keep_keys : ['image', 'label', 'length']
[2022/01/09 20:25:46] root INFO: loader :
[2022/01/09 20:25:46] root INFO: batch_size_per_card : 256
[2022/01/09 20:25:46] root INFO: drop_last : False
[2022/01/09 20:25:46] root INFO: num_workers : 8
[2022/01/09 20:25:46] root INFO: shuffle : False
[2022/01/09 20:25:46] root INFO: Global :
[2022/01/09 20:25:46] root INFO: cal_metric_during_train : True
[2022/01/09 20:25:46] root INFO: character_dict_path : /home/aistudio/训练集/amount/vocabulary.txt
[2022/01/09 20:25:46] root INFO: checkpoints : ./output/rec_chinese_common_v2.0/best_accuracy.pdparams
[2022/01/09 20:25:46] root INFO: debug : False
[2022/01/09 20:25:46] root INFO: distributed : False
[2022/01/09 20:25:46] root INFO: epoch_num : 500
[2022/01/09 20:25:46] root INFO: eval_batch_step : [100, 100]
[2022/01/09 20:25:46] root INFO: infer_img : doc/imgs_words/ch/word_1.jpg
[2022/01/09 20:25:46] root INFO: infer_mode : False
[2022/01/09 20:25:46] root INFO: log_smooth_window : 20
[2022/01/09 20:25:46] root INFO: max_text_length : 25
[2022/01/09 20:25:46] root INFO: pretrained_model : ./ch_ppocr_server_v2.0_rec_pre/best_accuracy
[2022/01/09 20:25:46] root INFO: print_batch_step : 10
[2022/01/09 20:25:46] root INFO: save_epoch_step : 3
[2022/01/09 20:25:46] root INFO: save_inference_dir : None
[2022/01/09 20:25:46] root INFO: save_model_dir : ./output/rec_chinese_common_v2.0
[2022/01/09 20:25:46] root INFO: save_res_path : ./output/rec/predicts_chinese_common_v2.0.txt
[2022/01/09 20:25:46] root INFO: use_gpu : True
[2022/01/09 20:25:46] root INFO: use_space_char : True
[2022/01/09 20:25:46] root INFO: use_visualdl : False
[2022/01/09 20:25:46] root INFO: Loss :
[2022/01/09 20:25:46] root INFO: name : CTCLoss
[2022/01/09 20:25:46] root INFO: Metric :
[2022/01/09 20:25:46] root INFO: main_indicator : acc
[2022/01/09 20:25:46] root INFO: name : RecMetric
[2022/01/09 20:25:46] root INFO: Optimizer :
[2022/01/09 20:25:46] root INFO: beta1 : 0.9
[2022/01/09 20:25:46] root INFO: beta2 : 0.999
[2022/01/09 20:25:46] root INFO: lr :
[2022/01/09 20:25:46] root INFO: learning_rate : 0.001
[2022/01/09 20:25:46] root INFO: name : Cosine
[2022/01/09 20:25:46] root INFO: warmup_epoch : 5
[2022/01/09 20:25:46] root INFO: name : Adam
[2022/01/09 20:25:46] root INFO: regularizer :
[2022/01/09 20:25:46] root INFO: factor : 4e-05
[2022/01/09 20:25:46] root INFO: name : L2
[2022/01/09 20:25:46] root INFO: PostProcess :
[2022/01/09 20:25:46] root INFO: name : CTCLabelDecode
[2022/01/09 20:25:46] root INFO: Train :
[2022/01/09 20:25:46] root INFO: dataset :
[2022/01/09 20:25:46] root INFO: data_dir : /home/aistudio/训练集/amount/images
[2022/01/09 20:25:46] root INFO: label_file_list : ['/home/aistudio/训练集/amount/train_list.txt']
[2022/01/09 20:25:46] root INFO: name : SimpleDataSet
[2022/01/09 20:25:46] root INFO: transforms :
[2022/01/09 20:25:46] root INFO: DecodeImage :
[2022/01/09 20:25:46] root INFO: channel_first : False
[2022/01/09 20:25:46] root INFO: img_mode : BGR
[2022/01/09 20:25:46] root INFO: RecAug : None
[2022/01/09 20:25:46] root INFO: CTCLabelEncode : None
[2022/01/09 20:25:46] root INFO: RecResizeImg :
[2022/01/09 20:25:46] root INFO: image_shape : [3, 32, 320]
[2022/01/09 20:25:46] root INFO: KeepKeys :
[2022/01/09 20:25:46] root INFO: keep_keys : ['image', 'label', 'length']
[2022/01/09 20:25:46] root INFO: loader :
[2022/01/09 20:25:46] root INFO: batch_size_per_card : 256
[2022/01/09 20:25:46] root INFO: drop_last : True
[2022/01/09 20:25:46] root INFO: num_workers : 8
[2022/01/09 20:25:46] root INFO: shuffle : True
[2022/01/09 20:25:46] root INFO: profiler_options : None
[2022/01/09 20:25:46] root INFO: train with paddle 2.0.2 and device CUDAPlace(0)
[2022/01/09 20:25:46] root INFO: Initialize indexs of datasets:['/home/aistudio/训练集/amount/val_list.txt']
W0109 20:25:46.365166 6137 device_context.cc:362] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 10.1, Runtime API Version: 10.1
W0109 20:25:46.370265 6137 device_context.cc:372] device: 0, cuDNN Version: 7.6.
[2022/01/09 20:25:52] root INFO: resume from ./output/rec_chinese_common_v2.0/best_accuracy
[2022/01/09 20:25:52] root INFO: metric in ckpt ***************
[2022/01/09 20:25:52] root INFO: acc:0.9699975750060625
[2022/01/09 20:25:52] root INFO: norm_edit_dis:0.9909632884925447
[2022/01/09 20:25:52] root INFO: fps:456.3268190347603
[2022/01/09 20:25:52] root INFO: best_epoch:36
[2022/01/09 20:25:52] root INFO: start_epoch:37
eval model:: 0%| | 0/2 [00:00<?, ?it/s]
eval model:: 50%|█████ | 1/2 [00:01<00:01, 1.72s/it]
eval model:: 100%|██████████| 2/2 [00:02<00:00, 1.29s/it]
eval model:: 100%|██████████| 2/2 [00:02<00:00, 1.11s/it]
[2022/01/09 20:25:54] root INFO: metric eval ***************
[2022/01/09 20:25:54] root INFO: acc:0.9699975750060625
[2022/01/09 20:25:54] root INFO: norm_edit_dis:0.9909632884925447
[2022/01/09 20:25:54] root INFO: fps:439.9205808964122
INFO 2022-01-09 20:25:56,825 launch.py:240] Local processes completed.
五、日期训练与评估
# 覆盖配置文件
!cp ./configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0_date.yml ~/rec_chinese_common_train_v2.0_date.yml
1.配置
Global:
use_gpu: true
epoch_num: 500
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/rec_chinese_common_v2.0_date
save_epoch_step: 3
# evaluation is run every 5000 iterations after the 4000th iteration
eval_batch_step: [100, 100]
cal_metric_during_train: True
pretrained_model: ./ch_ppocr_server_v2.0_rec_pre/best_accuracy
checkpoints:
save_inference_dir:
use_visualdl: False
infer_img: doc/imgs_words/ch/word_1.jpg
# for data or label process
character_dict_path: /home/aistudio/训练集/date/vocabulary.txt
max_text_length: 25
infer_mode: False
use_space_char: True
save_res_path: ./output/rec/predicts_chinese_common_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: 0.00004
Architecture:
model_type: rec
algorithm: CRNN
Transform:
Backbone:
name: ResNet
layers: 34
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 256
Head:
name: CTCHead
fc_decay: 0.00004
Loss:
name: CTCLoss
PostProcess:
name: CTCLabelDecode
Metric:
name: RecMetric
main_indicator: acc
Train:
dataset:
name: SimpleDataSet
data_dir: /home/aistudio/训练集/date/images
label_file_list: ["/home/aistudio/训练集/date/train_list.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/训练集/date/images
label_file_list: ["/home/aistudio/训练集/date/val_list.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
2.日期模型训练
# server模型
%cd ~/PaddleOCR/
!python tools/train.py -c ./configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0_date.yml
训练日志
[2022/01/09 21:15:41] root INFO: Initialize indexs of datasets:['/home/aistudio/训练集/date/train_list.txt']
[2022/01/09 21:15:56] root INFO: epoch: [69/500], iter: 410, lr: 0.000963, loss: 0.282951, acc: 0.978514, norm_edit_dis: 0.991211, reader_cost: 0.46780 s, batch_cost: 0.81243 s, samples: 1536, ips: 189.06251
[2022/01/09 21:16:06] root INFO: epoch: [69/500], iter: 413, lr: 0.000962, loss: 0.272608, acc: 0.979490, norm_edit_dis: 0.992269, reader_cost: 0.00150 s, batch_cost: 0.34571 s, samples: 1536, ips: 444.29798
[2022/01/09 21:16:10] root INFO: save model in ./output/rec_chinese_common_v2.0_date/latest
[2022/01/09 21:16:13] root INFO: save model in ./output/rec_chinese_common_v2.0_date/iter_epoch_69
[2022/01/09 21:16:13] root INFO: Initialize indexs of datasets:['/home/aistudio/训练集/date/train_list.txt']
[2022/01/09 21:16:39] root INFO: epoch: [70/500], iter: 419, lr: 0.000961, loss: 0.259662, acc: 0.983397, norm_edit_dis: 0.993701, reader_cost: 0.51926 s, batch_cost: 1.21329 s, samples: 3072, ips: 253.19623
[2022/01/09 21:16:43] root INFO: save model in ./output/rec_chinese_common_v2.0_date/latest
[2022/01/09 21:16:43] root INFO: Initialize indexs of datasets:['/home/aistudio/训练集/date/train_list.txt']
[2022/01/09 21:16:52] root INFO: epoch: [71/500], iter: 420, lr: 0.000961, loss: 0.259662, acc: 0.984373, norm_edit_dis: 0.994189, reader_cost: 0.55519 s, batch_cost: 0.67084 s, samples: 512, ips: 76.32220
[2022/01/09 21:17:09] root INFO: epoch: [71/500], iter: 425, lr: 0.000960, loss: 0.241655, acc: 0.985350, norm_edit_dis: 0.995117, reader_cost: 0.00232 s, batch_cost: 0.57741 s, samples: 2560, ips: 443.35555
3.日期模型评估
# server模型
!python -m paddle.distributed.launch tools/eval.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0_date.yml \
-o Global.checkpoints=./output/rec_chinese_common_v2.0_date/best_accuracy.pdparams
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
def convert_to_list(value, n, name, dtype=np.int):
----------- Configuration Arguments -----------
gpus: None
heter_worker_num: None
heter_workers:
http_port: None
ips: 127.0.0.1
log_dir: log
nproc_per_node: None
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_date.yml', '-o', 'Global.checkpoints=./output/rec_chinese_common_v2.0_date/best_accuracy.pdparams']
worker_num: None
workers:
------------------------------------------------
WARNING 2022-01-09 21:17:26,064 launch.py:316] Not found distinct arguments and compiled with cuda. Default use collective mode
launch train in GPU mode
INFO 2022-01-09 21:17:26,068 launch_utils.py:471] Local start 1 processes. First process distributed environment info (Only For Debug):
+=======================================================================================+
| Distributed Envs Value |
+---------------------------------------------------------------------------------------+
| PADDLE_TRAINER_ID 0 |
| PADDLE_CURRENT_ENDPOINT 127.0.0.1:50745 |
| PADDLE_TRAINERS_NUM 1 |
| PADDLE_TRAINER_ENDPOINTS 127.0.0.1:50745 |
| FLAGS_selected_gpus 0 |
+=======================================================================================+
INFO 2022-01-09 21:17:26,068 launch_utils.py:475] details abouts PADDLE_TRAINER_ENDPOINTS can be found in log/endpoints.log, and detail running logs maybe found in log/workerlog.0
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
def convert_to_list(value, n, name, dtype=np.int):
[2022/01/09 21:17:27] root INFO: Architecture :
[2022/01/09 21:17:27] root INFO: Backbone :
[2022/01/09 21:17:27] root INFO: layers : 34
[2022/01/09 21:17:27] root INFO: name : ResNet
[2022/01/09 21:17:27] root INFO: Head :
[2022/01/09 21:17:27] root INFO: fc_decay : 4e-05
[2022/01/09 21:17:27] root INFO: name : CTCHead
[2022/01/09 21:17:27] root INFO: Neck :
[2022/01/09 21:17:27] root INFO: encoder_type : rnn
[2022/01/09 21:17:27] root INFO: hidden_size : 256
[2022/01/09 21:17:27] root INFO: name : SequenceEncoder
[2022/01/09 21:17:27] root INFO: Transform : None
[2022/01/09 21:17:27] root INFO: algorithm : CRNN
[2022/01/09 21:17:27] root INFO: model_type : rec
[2022/01/09 21:17:27] root INFO: Eval :
[2022/01/09 21:17:27] root INFO: dataset :
[2022/01/09 21:17:27] root INFO: data_dir : /home/aistudio/训练集/date/images
[2022/01/09 21:17:27] root INFO: label_file_list : ['/home/aistudio/训练集/date/val_list.txt']
[2022/01/09 21:17:27] root INFO: name : SimpleDataSet
[2022/01/09 21:17:27] root INFO: transforms :
[2022/01/09 21:17:27] root INFO: DecodeImage :
[2022/01/09 21:17:27] root INFO: channel_first : False
[2022/01/09 21:17:27] root INFO: img_mode : BGR
[2022/01/09 21:17:27] root INFO: CTCLabelEncode : None
[2022/01/09 21:17:27] root INFO: RecResizeImg :
[2022/01/09 21:17:27] root INFO: image_shape : [3, 32, 320]
[2022/01/09 21:17:27] root INFO: KeepKeys :
[2022/01/09 21:17:27] root INFO: keep_keys : ['image', 'label', 'length']
[2022/01/09 21:17:27] root INFO: loader :
[2022/01/09 21:17:27] root INFO: batch_size_per_card : 256
[2022/01/09 21:17:27] root INFO: drop_last : False
[2022/01/09 21:17:27] root INFO: num_workers : 8
[2022/01/09 21:17:27] root INFO: shuffle : False
[2022/01/09 21:17:27] root INFO: Global :
[2022/01/09 21:17:27] root INFO: cal_metric_during_train : True
[2022/01/09 21:17:27] root INFO: character_dict_path : /home/aistudio/训练集/date/vocabulary.txt
[2022/01/09 21:17:27] root INFO: checkpoints : ./output/rec_chinese_common_v2.0_date/best_accuracy.pdparams
[2022/01/09 21:17:27] root INFO: debug : False
[2022/01/09 21:17:27] root INFO: distributed : False
[2022/01/09 21:17:27] root INFO: epoch_num : 500
[2022/01/09 21:17:27] root INFO: eval_batch_step : [100, 100]
[2022/01/09 21:17:27] root INFO: infer_img : doc/imgs_words/ch/word_1.jpg
[2022/01/09 21:17:27] root INFO: infer_mode : False
[2022/01/09 21:17:27] root INFO: log_smooth_window : 20
[2022/01/09 21:17:27] root INFO: max_text_length : 25
[2022/01/09 21:17:27] root INFO: pretrained_model : ./ch_ppocr_server_v2.0_rec_pre/best_accuracy
[2022/01/09 21:17:27] root INFO: print_batch_step : 10
[2022/01/09 21:17:27] root INFO: save_epoch_step : 3
[2022/01/09 21:17:27] root INFO: save_inference_dir : None
[2022/01/09 21:17:27] root INFO: save_model_dir : ./output/rec_chinese_common_v2.0_date
[2022/01/09 21:17:27] root INFO: save_res_path : ./output/rec/predicts_chinese_common_v2.0.txt
[2022/01/09 21:17:27] root INFO: use_gpu : True
[2022/01/09 21:17:27] root INFO: use_space_char : True
[2022/01/09 21:17:27] root INFO: use_visualdl : False
[2022/01/09 21:17:27] root INFO: Loss :
[2022/01/09 21:17:27] root INFO: name : CTCLoss
[2022/01/09 21:17:27] root INFO: Metric :
[2022/01/09 21:17:27] root INFO: main_indicator : acc
[2022/01/09 21:17:27] root INFO: name : RecMetric
[2022/01/09 21:17:27] root INFO: Optimizer :
[2022/01/09 21:17:27] root INFO: beta1 : 0.9
[2022/01/09 21:17:27] root INFO: beta2 : 0.999
[2022/01/09 21:17:27] root INFO: lr :
[2022/01/09 21:17:27] root INFO: learning_rate : 0.001
[2022/01/09 21:17:27] root INFO: name : Cosine
[2022/01/09 21:17:27] root INFO: warmup_epoch : 5
[2022/01/09 21:17:27] root INFO: name : Adam
[2022/01/09 21:17:27] root INFO: regularizer :
[2022/01/09 21:17:27] root INFO: factor : 4e-05
[2022/01/09 21:17:27] root INFO: name : L2
[2022/01/09 21:17:27] root INFO: PostProcess :
[2022/01/09 21:17:27] root INFO: name : CTCLabelDecode
[2022/01/09 21:17:27] root INFO: Train :
[2022/01/09 21:17:27] root INFO: dataset :
[2022/01/09 21:17:27] root INFO: data_dir : /home/aistudio/训练集/date/images
[2022/01/09 21:17:27] root INFO: label_file_list : ['/home/aistudio/训练集/date/train_list.txt']
[2022/01/09 21:17:27] root INFO: name : SimpleDataSet
[2022/01/09 21:17:27] root INFO: transforms :
[2022/01/09 21:17:27] root INFO: DecodeImage :
[2022/01/09 21:17:27] root INFO: channel_first : False
[2022/01/09 21:17:27] root INFO: img_mode : BGR
[2022/01/09 21:17:27] root INFO: RecAug : None
[2022/01/09 21:17:27] root INFO: CTCLabelEncode : None
[2022/01/09 21:17:27] root INFO: RecResizeImg :
[2022/01/09 21:17:27] root INFO: image_shape : [3, 32, 320]
[2022/01/09 21:17:27] root INFO: KeepKeys :
[2022/01/09 21:17:27] root INFO: keep_keys : ['image', 'label', 'length']
[2022/01/09 21:17:27] root INFO: loader :
[2022/01/09 21:17:27] root INFO: batch_size_per_card : 512
[2022/01/09 21:17:27] root INFO: drop_last : True
[2022/01/09 21:17:27] root INFO: num_workers : 8
[2022/01/09 21:17:27] root INFO: shuffle : True
[2022/01/09 21:17:27] root INFO: profiler_options : None
[2022/01/09 21:17:27] root INFO: train with paddle 2.0.2 and device CUDAPlace(0)
[2022/01/09 21:17:27] root INFO: Initialize indexs of datasets:['/home/aistudio/训练集/date/val_list.txt']
W0109 21:17:27.794884 15547 device_context.cc:362] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 10.1, Runtime API Version: 10.1
W0109 21:17:27.800077 15547 device_context.cc:372] device: 0, cuDNN Version: 7.6.
[2022/01/09 21:17:33] root INFO: resume from ./output/rec_chinese_common_v2.0_date/best_accuracy
[2022/01/09 21:17:33] root INFO: metric in ckpt ***************
[2022/01/09 21:17:33] root INFO: acc:0.991173555371896
[2022/01/09 21:17:33] root INFO: norm_edit_dis:0.9953431509515168
[2022/01/09 21:17:33] root INFO: fps:443.3070479891199
[2022/01/09 21:17:33] root INFO: best_epoch:67
[2022/01/09 21:17:33] root INFO: start_epoch:68
eval model:: 0%| | 0/2 [00:00<?, ?it/s]
eval model:: 50%|█████ | 1/2 [00:01<00:01, 1.59s/it]
eval model:: 100%|██████████| 2/2 [00:01<00:00, 1.17s/it]
eval model:: 100%|██████████| 2/2 [00:01<00:00, 1.02it/s]
[2022/01/09 21:17:35] root INFO: metric eval ***************
[2022/01/09 21:17:35] root INFO: acc:0.991173555371896
[2022/01/09 21:17:35] root INFO: norm_edit_dis:0.9953431509515168
[2022/01/09 21:17:35] root INFO: fps:441.1924785284007
INFO 2022-01-09 21:17:38,107 launch.py:240] Local processes completed.
六、结果预测
修改 tools/infer_rec.py
with open(save_res_path, "w") as fout:
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(os.path.basename(file) + "\t" + post_result[0][0]+"\n")
logger.info("success!")
!cp ~/infer_rec.py ~/PaddleOCR/tools/infer_rec.py
1.金额预测
# 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/测试集/amount/images" \
Global.checkpoints=./output/rec_chinese_common_v2.0/best_accuracy
输出日志
[2022/01/09 21:33:25] root INFO: infer_img: /home/aistudio/测试集/amount/images/8bb1941c760a2c1d01762c63a2220571.jpg
[2022/01/09 21:33:25] root INFO: result: 壹仟捌佰贰拾肆元贰角叁分 0.974654
[2022/01/09 21:33:25] root INFO: infer_img: /home/aistudio/测试集/amount/images/8bb1941c760a2c1d01762cc9ce6e6029.jpg
[2022/01/09 21:33:25] root INFO: result: 壹万叁仟元整 0.9725123
[2022/01/09 21:33:25] root INFO: infer_img: /home/aistudio/测试集/amount/images/8bb1941c760a2c1d017640031a39469f.jpg
[2022/01/09 21:33:25] root INFO: result: 玖万柒仟陆佰伍拾柒元伍角肆分 0.9835057
[2022/01/09 21:33:25] root INFO: infer_img: /home/aistudio/测试集/amount/images/8bb1941c760a2c1d0176415a9ec807fe.jpg
[2022/01/09 21:33:25] root INFO: result: 伍万柒仟元整 0.9003706
[2022/01/09 21:33:25] root INFO: infer_img: /home/aistudio/测试集/amount/images/8bb1941c760a2c1d017645417b3b5c5e.jpg
[2022/01/09 21:33:25] root INFO: result: 叁万叁仟陆佰肆拾伍元肆角玖分 0.9679618
[2022/01/09 21:33:25] root INFO: infer_img: /home/aistudio/测试集/amount/images/8bb1941c760a2c1d017646019cf7387d.jpg
[2022/01/09 21:33:25] root INFO: result: 伍仟零伍拾玖元肆角整 0.98454416
[2022/01/09 21:33:25] root INFO: infer_img: /home/aistudio/测试集/amount/images/8bb1941c760a2c1d01764b77a53735a0.jpg
[2022/01/09 21:33:25] root INFO: result: 柒佰肆拾壹元柒角叁分 0.9320029
!head ./output/rec/predicts_chinese_common_v2.0.txt
8bb1941c760a2c1d017626c361da6c4d.jpg 壹万伍仟叁佰柒拾元正
8bb1941c760a2c1d01762b943a624421.jpg 壹拾捌万伍仟元整
8bb1941c760a2c1d01762c63a2220571.jpg 壹仟捌佰贰拾肆元贰角叁分
8bb1941c760a2c1d01762cc9ce6e6029.jpg 壹万叁仟元整
8bb1941c760a2c1d017640031a39469f.jpg 玖万柒仟陆佰伍拾柒元伍角肆分
8bb1941c760a2c1d0176415a9ec807fe.jpg 伍万柒仟元整
8bb1941c760a2c1d017645417b3b5c5e.jpg 叁万叁仟陆佰肆拾伍元肆角玖分
8bb1941c760a2c1d017646019cf7387d.jpg 伍仟零伍拾玖元肆角整
8bb1941c760a2c1d01764b77a53735a0.jpg 柒佰肆拾壹元柒角叁分
8bb1941c7657bb0101765edc72b01d52.jpg 壹万贰仟贰佰柒拾肆元整
2.日期预测
# server模型
!python tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0_date.yml \
-o Global.infer_img="/home/aistudio/测试集/date/images" \
Global.checkpoints=./output/rec_chinese_common_v2.0_date/best_accuracy
输出日志
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248693017a421e33087817.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.94455576
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248693017a5653c6d32277.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.9286475
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248693017a56bc0a886d10.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.9824974
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248693017a8955798e0c20.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.9235918
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248693017aa87d11e203b6.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.97518396
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248694017a31b07a3a1b2c.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.9815266
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248694017a3cfacc781d85.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.9835255
[2022/01/09 21:34:48] root INFO: infer_img: /home/aistudio/测试集/date/images/0_8bb1941c7a248694017a987384ab04e8.jpg
[2022/01/09 21:34:48] root INFO: result: 贰零贰壹 0.9853095
!head ./output/rec/predicts_chinese_common_v2.0_date.txt
0_8bb1941c7a248693017a377ec36606b7.jpg 贰零贰壹
0_8bb1941c7a248693017a421e33087817.jpg 贰零贰壹
0_8bb1941c7a248693017a5653c6d32277.jpg 贰零贰壹
0_8bb1941c7a248693017a56bc0a886d10.jpg 贰零贰壹
0_8bb1941c7a248693017a8955798e0c20.jpg 贰零贰壹
0_8bb1941c7a248693017aa87d11e203b6.jpg 贰零贰壹
0_8bb1941c7a248694017a31b07a3a1b2c.jpg 贰零贰壹
0_8bb1941c7a248694017a3cfacc781d85.jpg 贰零贰壹
0_8bb1941c7a248694017a987384ab04e8.jpg 贰零贰壹
0_8bb1941c7a248694017a9ee81daf390b.jpg 贰零贰壹
.jpg 贰零贰壹
0_8bb1941c7a248694017a3cfacc781d85.jpg 贰零贰壹
0_8bb1941c7a248694017a987384ab04e8.jpg 贰零贰壹
0_8bb1941c7a248694017a9ee81daf390b.jpg 贰零贰壹
合并 ./output/rec/predicts_chinese_common_v2.0.txt 与 /output/rec/predicts_chinese_common_v2.0_date.txt ，并提交即可取得成绩。
目前该比赛已经关闭。
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)