Thanks to Li Liang's help. The best result of my pytorch model is 0.783 ODS F-score now.
The result of my pytorch model will be released in the future
| Method | ODS F-score on BSDS500 dataset |
|---|---|
| RCF-PointRend | 0.807 |
| RCF | 0.806 |
Install pytorch. The code is tested under 11.6 cuda version and Python 3.9 on Ubuntu 20.04. There are also some dependencies for a few Python libraries for data processing and visualizations like cv2 etc. It's highly recommended that you have access to GPUs.
To train a RCF-PointRend model on BSDS500:
python train_RCF_PointRend.py
To resume the training on BSDS500:
python train_RCF_PointRend.py --checkpoint '[.pth file]'
If you have multiple GPUs on your machine, you can also run the multi-GPU version training:
CUDA_VISIBLE_DEVICES=0,1 python train_multi_gpu.py --num_gpus 2
- To download the pretrained model. Please click https://drive.google.com/open?id=1TupHeoBKawrniDka0Hc64m3BG4OKG8nM (This pretrained model is not the best model, just for communicating)
- To download the vgg16 pretrained model which is used for the backbone. Please click https://drive.google.com/file/d/1lUhPKKj-BSOH7yQL0mOIavvrUbjydPp5/view?usp=sharing.
- To download the checkpoint of training epoch 14. Please ckick https://pan.baidu.com/s/1LS9CF5Zrm4gBj9p-HqJDSQ (Access Code: rkq8).
- To download the HED-BSDS dataset. Please click https://pan.baidu.com/s/13be7JVWzbup_h-04axRrSw (Access Code: 2nzk) or https://vcl.ucsd.edu/hed/HED-BSDS.tar.
- To download the BSDS500 dataset. Please click https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html.
- To evaluate the model, you can learn more information in https://www.jianshu.com/p/eda277063867.