153 lines
4.4 KiB
Markdown
153 lines
4.4 KiB
Markdown
# PaddleOCR
|
||
|
||
------
|
||
|
||
## 数据集
|
||
|
||
该部分内容均在PPOCRLabel目录下进行
|
||
|
||
```bash
|
||
# 进入PPOCRLabel目录
|
||
cd .\PPOCRLabel\
|
||
```
|
||
|
||
### 打标
|
||
|
||
可以对PPOCRLabel.py直接使用PyCharm中的Run，但是默认是英文的
|
||
|
||
```bash
|
||
# 以中文运行打标应用
|
||
python PPOCRLabel.py --lang ch
|
||
# 含有关键词提取的打标
|
||
python PPOCRLabel.py --lang ch --kie True
|
||
```
|
||
|
||
### 划分数据集
|
||
|
||
```bash
|
||
python gen_ocr_train_val_test.py --trainValTestRatio 6:2:2 --datasetRootPath ../train_data/drivingData
|
||
```
|
||
|
||
------
|
||
|
||
## 检测模型
|
||
|
||
先回到项目根目录
|
||
|
||
### 训练
|
||
|
||
```bash
|
||
python tools/train.py -c configs/det/ch_PP-OCRv4/ch_PP-OCRv4_det_student.yml
|
||
```
|
||
|
||
### 测试
|
||
|
||
```bash
|
||
python tools/infer_det.py -c configs/det/ch_PP-OCRv4/ch_PP-OCRv4_det_student.yml -o Global.pretrained_model=output/det_v4_bankcard/best_accuracy.pdparams Global.infer_img=train_data/drivingData/1.jpg
|
||
```
|
||
|
||
### 恢复训练
|
||
|
||
```bash
|
||
python tools/train.py -c configs/det/ch_PP-OCRv4/ch_PP-OCRv4_det_student.yml -o Global.checkpoints=./output/det_v4_bankcard/latest
|
||
```
|
||
|
||
------
|
||
|
||
## 识别模型
|
||
|
||
### 训练
|
||
|
||
```bash
|
||
python tools/train.py -c configs/rec/PP-OCRv4/ch_PP-OCRv4_rec_ampO2_ultra.yml
|
||
```
|
||
|
||
### 测试
|
||
|
||
```bash
|
||
python tools/infer_rec.py -c configs/rec/PP-OCRv4/ch_PP-OCRv4_rec_ampO2_ultra.yml -o Global.pretrained_model=output/rec_v4_bankcard/best_accuracy.pdparams Global.infer_img=train_data/drivingData/crop_img/1_crop_0.jpg
|
||
```
|
||
|
||
------
|
||
|
||
## 推理模型
|
||
|
||
### 检测模型转换
|
||
|
||
```bash
|
||
python tools/export_model.py -c configs/det/ch_PP-OCRv4/ch_PP-OCRv4_det_student.yml -o Global.pretrained_model=output/det_v4_bankcard/best_accuracy.pdparams
|
||
```
|
||
|
||
### 识别模型转换
|
||
|
||
```bash
|
||
python tools/export_model.py -c configs/rec/PP-OCRv4/ch_PP-OCRv4_rec_ampO2_ultra.yml -o Global.pretrained_model=output/rec_v4_bankcard/best_accuracy.pdparams
|
||
```
|
||
|
||
### 检测识别测试
|
||
|
||
```bash
|
||
python tools/infer/predict_system.py --det_model_dir=inference_model/det_v4_bankcard --rec_model_dir=inference_model/rec_v4_bankcard --rec_char_dict_path=ppocr/utils/num_dict.txt --image_dir=train_data/drivingData/1.jpg
|
||
```
|
||
|
||
------
|
||
|
||
## 移动端模型
|
||
|
||
### 检测模型转换
|
||
|
||
```bash
|
||
paddle_lite_opt --model_file=inference_model/det_v4_bankcard/inference.pdmodel --param_file=inference_model/det_v4_bankcard/inference.pdiparams --optimize_out=inference_model/det_v4_nb_bankcard --valid_targets=arm --optimize_out_type=naive_buffer
|
||
```
|
||
|
||
### 识别模型转换
|
||
|
||
```bash
|
||
paddle_lite_opt --model_file=inference_model/rec_v4_bankcard/inference.pdmodel --param_file=inference_model/rec_v4_bankcard/inference.pdiparams --optimize_out=inference_model/rec_v4_nb_bankcard --valid_targets=arm --optimize_out_type=naive_buffer
|
||
```
|
||
|
||
------
|
||
|
||
------
|
||
|
||
# PaddleNLP
|
||
|
||
## 数据集
|
||
|
||
使用Label Studio进行数据标注，安装过程省略
|
||
|
||
```bash
|
||
# 打开Anaconda Prompt
|
||
# 激活安装Label Studio的环境
|
||
conda activate label-studio
|
||
# 启动Label Studio
|
||
label-studio start
|
||
```
|
||
|
||
[打标流程](https://github.com/PaddlePaddle/PaddleNLP/blob/develop/applications/information_extraction/label_studio_doc.md)
|
||
|
||
### 数据转换
|
||
|
||
```bash
|
||
# 进入PaddleNLP\applications\information_extraction后执行
|
||
python label_studio.py --label_studio_file ./document/data/label_studio.json --save_dir ./document/data --splits 0.8 0.1 0.1 --task_type ext
|
||
```
|
||
|
||
|
||
|
||
------
|
||
|
||
## 训练模型
|
||
|
||
```bash
|
||
# 进入PaddleNLP\applications\information_extraction\document后执行(双卡训练)
|
||
python -u -m paddle.distributed.launch --gpus "0,1" finetune.py --device gpu --logging_steps 5 --save_steps 25 --eval_steps 25 --seed 42 --model_name_or_path uie-x-base --output_dir ./checkpoint/model_best --train_path data/train.txt --dev_path data/dev.txt --max_seq_len 512 --per_device_train_batch_size 8 --per_device_eval_batch_size 8 --num_train_epochs 10 --learning_rate 1e-5 --do_train --do_eval --do_export --export_model_dir ./checkpoint/model_best --overwrite_output_dir --disable_tqdm False --metric_for_best_model eval_f1 --load_best_model_at_end True --save_total_limit 1
|
||
```
|
||
|
||
------
|
||
|
||
参考：
|
||
|
||
[PaddleOCR训练属于自己的模型详细教程](https://blog.csdn.net/qq_52852432/article/details/131817619?utm_medium=distribute.pc_relevant.none-task-blog-2~default~baidujs_baidulandingword~default-0-131817619-blog-124628731.235^v40^pc_relevant_3m_sort_dl_base1&spm=1001.2101.3001.4242.1&utm_relevant_index=3)
|
||
[端侧部署](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.7/deploy/lite/readme_ch.md)
|
||
[PaddleNLP关键信息抽取](https://blog.csdn.net/z5z5z5z56/article/details/130346646) |