Vitali Petsiuk

Fair point, I will update the repository soon. Meanwhile, kindly refer to the section 3.2 Mask generation of the paper. https://arxiv.org/pdf/1806.07421.pdf N is the number of masks used, (s, s)...

Hi @ytsheng, Looks good, thanks for your work. I will try to merge and adapt it by the end of the week.

Hi @wenouyang. Yes, current implementation requires model to be rebuild with different `input_shape`. Try changing `img_w`, `img_h` in build_model.py. I will look into support of different input shapes later.

@codepujan The provided weights are not trained. I will look into changing them. You can also check https://github.com/imlab-uiip/lung-segmentation-2d.

Hi, @opnumten. In order to implement index based upsampling with tensorflow you would need to modify tf pooling function so that it would memorize max indices, similarly you would need...

Yes, getting meaningful saliency maps for models with bad accuracy/recall can be challenging. I don't see a problem with using low occlusion probability -- the intuition of the approach still...