This link is for person which is new to OCR CSDN Blog
Recently updated:
- 2020.12.22 Update CRNN+CTCLoss + CenterLoss training
- 2020.09.18 Update text detection documentation
- 2020.09.12 Update DB,pse,pan,sast,crnn Training test code and pre-trained model
Currently completed:
- DBnet Paper link
- PSEnet Paper link
- PANnet Paper link
- SASTnet Paper link
- CRNN Paper link
Next plan：
- Model transfer onnx and test
- Model compression (pruning)
- Model compression (quantization)
- Model distillation
- Deploy tensorrt
- Training generalized OCR model
- Deploy with chinese_lite
- Mobile deployment
The training is only on the ICDAR2015 text detection public data set, and the algorithm result is as follows:
| Model | Backbone Network | Precision | Recall | Hmean | Download Link |
|---|---|---|---|---|---|
| DB | ResNet50_7*7 | 85.88% | 79.10% | 82.35% | Download link(code:fxw6) |
| DB | ResNet50_3*3 | 86.51% | 80.59% | 83.44% | Download link(code:fxw6) |
| DB | MobileNetV3 | 82.89% | 75.83% | 79.20% | Download link(code:fxw6) |
| SAST | ResNet50_7*7 | 85.72% | 78.38% | 81.89% | Download link(code:fxw6) |
| SAST | ResNet50_3*3 | 86.67% | 76.74% | 81.40% | Download link(code:fxw6) |
| PSE | ResNet50_7*7 | 84.10% | 80.01% | 82.01% | Download link(code:fxw6) |
| PSE | ResNet50_3*3 | 82.56% | 78.91% | 80.69% | Download link(code:fxw6) |
| PAN | ResNet18_7*7 | 81.80% | 77.08% | 79.37% | Download link(code:fxw6) |
| PAN | ResNet18_3*3 | 83.78% | 75.15% | 79.23% | Download link(code:fxw6) |
Here, mobilev3 is used as the backbone. As a result of testing on icdar2015, the initial size of the uncompressed model is 2.4M.
- Compress the backbone
| Model | Pruned method | Ratio | Model size(M) | Precision | Recall | Hmean |
|---|---|---|---|---|---|---|
| DB | no | 0 | 2.4 | 84.04% | 75.34% | 79.46% |
| DB | backbone | 0.5 | 1.9 | 83.74% | 73.18% | 78.10% |
| DB | backbone | 0.6 | 1.58 | 84.46% | 69.90% | 76.50% |
- Compress the entire model
| Model | pruned method | ratio | model size(M) | precision | recall | Hmean |
|---|---|---|---|---|---|---|
| DB | no | 0 | 2.4 | 85.70% | 74.77% | 79.86% |
| DB | total | 0.6 | 1.42 | 82.97% | 75.10% | 78.84% |
| DB | total | 0.65 | 1.15 | 85.14% | 72.84% | 78.51% |
| Model | Teacher | Student | Model size(M) | Precision | Recall | Hmean | Improve(%) |
|---|---|---|---|---|---|---|---|
| DB | no | mobilev3 | 2.4 | 85.70% | 74.77% | 79.86% | - |
| DB | resnet50 | mobilev3 | 2.4 | 86.37% | 77.22% | 81.54% | 1.68 |
| DB | no | mobilev3 | 1.42 | 82.97% | 75.10% | 78.84% | - |
| DB | resnet50 | mobilev3 | 1.42 | 85.88% | 76.16% | 80.73% | 1.89 |
| DB | no | mobilev3 | 1.15 | 85.14% | 72.84% | 78.51% | - |
| DB | resnet50 | mobilev3 | 1.15 | 85.60% | 74.72% | 79.79% | 1.28 |
WeChat number: -fxwispig-
- https://github.com/PaddlePaddle/PaddleOCR
- https://github.com/whai362/PSENet
- https://github.com/whai362/pan_pp.pytorch
- https://github.com/WenmuZhou/PAN.pytorch
- https://github.com/xiaolai-sqlai/mobilenetv3
- https://github.com/BADBADBADBOY/DBnet-lite.pytorch
- https://github.com/BADBADBADBOY/Psenet_v2
- https://github.com/BADBADBADBOY/pse-lite.pytorch