Scene Text Detection and Recognition is an open source scene text detection and recognition benchmark based on PyTorch
- Various backbones and pretrained models
- Large-scale training configs
- High efficiency and extensibility
This is my experiment eviroument
- python3.6
- pytorch1.6.0+cu101
- implement DB
- implement CRNN
- implement ABCNet
Supported:
- ICDAR15
- ICDAR17MLT
- sythtext800K
- Total Text
- CTW1500
- 2019ArT
Supported text detection:
- EAST EAST: An Efficient and Accurate Scene Text Detector
- Psenet Shape Robust Text Detection with Progressive Scale Expansion Network
- DB Real-time Scene Text Detection with Differentiable Binarization
Supported text recognition:
Supported End to End:
Please refer to install.md for installation and dataset preparation.
All models are trained in the same condition, and might not get the best result
Method | Backbone | Pretrain | Resolution | Dataset | Precision | Recall | F-score | FPS |
---|---|---|---|---|---|---|---|---|
EAST | VGG16 | - | 512 | ICDAR15 | 0.81 | 0.81 | 0.81 | - |
EAST | VGG16 | - | 512 | ICDAR17 | 0.72 | 0.61 | 0.66 | - |
EAST | VGG16 | SynthText | 512 | ICDAR15 | 0.82 | 0.824 | 0.822 | - |
Method | Backbone | Pretrain | Resolution | Dataset | Precision | Recall | F-score | FPS |
---|---|---|---|---|---|---|---|---|
PSENet(1s) | ResNet50 | - | 640*640 | SynthText | - | - | - | - |
PSENet(1s) | ResNet50 | - | 640*640 | ICDAR15 | 0.816 | 0.795 | 0.805 | - |
PSENet(1s) | ResNet50 | - | 640*640 | ICDAR17(val) | 0.755 | 0.614 | 0.677 | - |
PSENet(1s) | ResNet50 | - | 640*640 | ICDAR17(test) | 0.762 | 0.643 | 0.698 | - |
PSENet(1s) | ResNet50 | - | 640*640 | Total-Text | 0.8255 | 0.7597 | 0.7913 | 3.0 |
PSENet(1s) | ResNet50 | SynthText | 640*640 | ICDAR15 | 0.864 | 0.835 | 0.850 | - |
PSENet(1s) | ResNet50 | SynthText+ICDAR17 | 640*640 | ICDAR15 | 0.883 | 0.853 | 0.868 | 3.0 |
PSENet(1s) | ResNet50 | SynthText | 640*640 | Total-Text | 0.834 | 0.781 | 0.807 | 3.0 |
EAST: https://github.com/SakuraRiven/EAST
PSENet: https://github.com/WenmuZhou/PSENet.pytorch
DB: https://github.com/WenmuZhou/DBNet.pytorch
https://github.com/WenmuZhou/PytorchOCR
Eamil: [email protected]