Skip to content

Commit 4c09232

Browse files
frotmsfrotms
committedJan 10, 2021
update
1 parent 73b080e commit 4c09232

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

101 files changed

+10734
-2
lines changed
 

‎README.md

+149-2
Original file line numberDiff line numberDiff line change
@@ -1,2 +1,149 @@
1-
# PaddleOCR2Pytorch
2-
Converted from [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)
1+
# [PaddleOCR2Pytorch](https://github.com/frotms/PaddleOCR2Pytorch)
2+
3+
## 简介
4+
5+
**”真·白嫖“**[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)
6+
7+
## 注意
8+
9+
`PytorchOCR`由`PaddleOCR-2.0rc1+`动态图版本移植。
10+
11+
## 特性
12+
13+
高质量推理模型，准确的识别效果
14+
15+
- 超轻量ptocr_mobile移动端系列
16+
- 通用ptocr_server系列
17+
- 支持中英文数字组合识别、竖排文本识别、长文本识别
18+
19+
## 模型列表
20+
21+
`.pth`模型下载链接：https://pan.baidu.com/s/1r1DELT8BlgxeOP2RqREJEg 提取码：6clx
22+
23+
### 检测模型
24+
25+
| 模型名称 | PaddleOCR对应模型 |
26+
| ------------------------------------ | ------------------------------------------------------------ |
27+
| `ch_ptocr_mobile_v2.0_det_infer.pth` | ch_ppocr_mobile_v2.0_det: [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar) |
28+
| `ch_ptocr_server_v2.0_det_infer.pth` | ch_ppocr_server_v2.0_det: [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar) |
29+
30+
### 识别模型
31+
32+
| 模型名称 | PaddleOCR对应模型 |
33+
| ------------------------------------ | ------------------------------------------------------------ |
34+
| `ch_ptocr_mobile_v2.0_rec_infer.pth` | ch_ppocr_mobile_v2.0_rec: [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) |
35+
| `ch_ptocr_server_v2.0_rec_infer.pth` | ch_ppocr_server_v2.0_rec: [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar) |
36+
37+
### 分类模型
38+
39+
| 模型名称 | PaddleOCR对应模型 |
40+
| ------------------------------------ | ------------------------------------------------------------ |
41+
| `ch_ptocr_mobile_v2.0_cls_infer.pth` | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |
42+
43+
## 效果展示
44+
45+
- 中文模型
46+
47+
<div align="center">
48+
<img src="./doc/imgs_results/ch_ptocr_mobile_v2.0/11.jpg" width="800">
49+
<img src="./doc/imgs_results/ch_ptocr_mobile_v2.0/00015504.jpg" width="800">
50+
<img src="./doc/imgs_results/ch_ptocr_mobile_v2.0/00056221.jpg" width="800">
51+
<img src="./doc/imgs_results/ch_ptocr_mobile_v2.0/1.jpg" width="800">
52+
</div>
53+
54+
55+
56+
## Pipeline
57+
58+
<div align="center">
59+
<img src="./doc/framework.png" width="800">
60+
</div>
61+
62+
PP-OCR是一个实用的超轻量OCR系统。主要由DB文本检测、检测框矫正和CRNN文本识别三部分组成。该系统从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941 。其中FPGM裁剪器和PACT量化的实现可以参考[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)。
63+
64+
## 推理
65+
66+
### 环境
67+
68+
```
69+
shapely
70+
numpy
71+
pillow
72+
pyclipper
73+
opencv-python <= 4.2.0.32
74+
pytorch
75+
```
76+
77+
### 中文检测模型推理
78+
79+
```
80+
python3 ./tools/infer/predict_det.py --image_dir ./doc/imgs --model_path your_det_pth_path.pth
81+
```
82+
83+
### 中文识别模型推理
84+
85+
```
86+
python3 ./tools/infer/predict_rec.py --image_dir ./doc/imgs_words --model_path your_rec_pth_path.pth
87+
```
88+
89+
### 中文方向分类模型推理
90+
91+
```
92+
python3 ./tools/infer/predict_cls.py --image_dir ./doc/imgs_words --model_path your_cls_pth_path.pth
93+
```
94+
95+
### 文本检测、方向分类和文字识别串联推理
96+
97+
#### 使用方向分类器
98+
99+
```
100+
python3 ./tools/infer/predict_system.py --image_dir ./doc/imgs --det_model_path your_det_pth_path.pth --rec_model_path your_rec_pth_path.pth --use_angle_cls --cls_model_path your_cls_pth_path.pth
101+
```
102+
103+
#### 不适用方向分类器
104+
105+
```
106+
python3 ./tools/infer/predict_system.py --image_dir ./doc/imgs --det_model_path your_det_pth_path.pth --rec_model_path your_rec_pth_path.pth
107+
```
108+
109+
## PaddleOCR2Pytorch
110+
111+
### 环境
112+
113+
```
114+
shapely
115+
numpy
116+
pillow
117+
pyclipper
118+
opencv-python <= 4.2.0.32
119+
pytorch
120+
paddlepaddle==2.0rc1
121+
```
122+
123+
### 模型转换
124+
125+
模型路径详见**PaddleOCR对应模型**或者**百度网盘链接**：https://pan.baidu.com/s/1getAprT2l_JqwhjwML0g9g
126+
提取码：lmv7
127+
128+
```
129+
python3 ./converter/ch_ppocr_mobile_v2.0_det_converter.py --src_model_path paddle_ch_ppocr_mobile_v2.0_det_dir
130+
131+
python3 ./converter/ch_ppocr_server_v2.0_det_converter.py --src_model_path paddle_ch_ppocr_server_v2.0_det_dir
132+
133+
python3 ./converter/ch_ppocr_mobile_v2.0_rec_converter.py --src_model_path paddle_ch_ppocr_mobile_v2.0_rec_dir
134+
135+
python3 ./converter/ch_ppocr_server_v2.0_rec_converter.py --src_model_path paddle_ch_ppocr_server_v2.0_rec_dir
136+
137+
python3 ./converter/ch_ppocr_mobile_v2.0_cls_converter.py --src_model_path paddle_ch_ppocr_mobile_v2.0_cls_dir
138+
```
139+
140+
## [FAQ](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.0-rc1-0/doc/doc_ch/FAQ.md)
141+
142+
## 参考
143+
144+
- [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)
145+
- [PytorchOCR](https://github.com/WenmuZhou/PytorchOCR)
146+
- [Paddle](https://github.com/PaddlePaddle)
147+
- [Pytorch](https://pytorch.org/)
148+
- [https://github.com/frotms/image_classification_pytorch](https://github.com/frotms/image_classification_pytorch)
149+
- [https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.0-rc1-0/doc/doc_ch/models_list.md](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.0-rc1-0/doc/doc_ch/models_list.md)

‎__init__.py

Whitespace-only changes.

0 commit comments

Comments
 (0)
Please sign in to comment.