This code is based on PixlLink and PSENet, the performance is not satisfactory
- Python 3.6
- PyTorch v0.4.1+
- opencv-python 3.4
CUDA_VISIBLE_DEVICES=0,1,2,3 python train_ic15.py --arch vgg16 --batch_size 24
CUDA_VISIBLE_DEVICES=0 python test_ic15.py --resume [path of model]
cd eval
sh eval_ic15.sh
Dataset | Pretrained | Precision (%) | Recall (%) | F-measure (%) | FPS (1080Ti) | Input |
---|---|---|---|---|---|---|
ICDAR2015 | No | 81.2 | 75 | 78 | 5 | 1280*768 |
- Find the bug of the performance issue.
- Accomplish the code with better config file and more datasets