Update the save_visualization function.

SamsungLabs / fbrs_interactive_segmentation

[CVPR2020] f-BRS: Rethinking Backpropagating Refinement for Interactive Segmentation https://arxiv.org/abs/2001.10331

Mozilla Public License 2.0

583 stars 94 forks source link

Update the save_visualization function. #21

Closed TwistedW closed 4 years ago

TwistedW commented 4 years ago

Thank you for providing such beautiful code. Make some suggestions based on personal use. It can be seen from the code that the predicted_mask is saved in the form of a heat map. Your team adopted cv2.COLORMAP_HOT. maphot But after np.hstack, result[:, :, ::-1] is added. This should be used in the processing of the input image, not in the predicted_mask. The result before the change is like this. 125200_instance_segmentation I made some adjustments.The results obtained are feasible.

ptrvilya commented 4 years ago

Hi! Thanks for your interest in our work. Reversing the whole tensor with image and predicted mask here is intentional, because for us blue color map seems better than the red one. We also use the same colors in our AdaptIS repository. So to keep things consistent, I suggest that we keep our current implementation as is.