M2CS Pytorch

pytorch+cuda11.2

Datasets

• test_data: The data for testing.
  • CUHK: Contains 100 testing images of CUHK Dataset and it's GT. Download here
  • DUT: Contains 500 testing images of DUT Dataset and it's GT. Download here

Test

You can use the following command to test：

python test.py --stict PRETRAINED_WEIGHT --image_path IMG_PATH --mask_save_path SAVE_PATH

For example:

python test.py --stict generator_pretrained.pth --image_path CUHK/xu100-source/ --mask_save_path result

We provide a pretrained model for testing.

Train

Since our article is being reviewed, it is not suitable to publish the training code now. If the paper is accepted, we promise to upload the training code at the first time.

Eval

In our paper, our method was compared with different methods, the output of these methods could be download here

You can use the following command to F-measure, MAE and Accuracy：

python eval.py --gt_path --mask_path

You can run S-measure-code/demo2.m to get S-measure.