Instance Segmentation on the Whole Image

[zhaoyangwei123]

@LiWentomng In your paper, you mentioned restricting the level set to gt bounding box for segmentation and also mentioned that the training result is poor on the whole image. How can I get the boxlevelset to run on the whole image?

[LiWentomng]

Hello, @zhaoyangwei123 The gt bounding box can prevent the interference from background and other objects. We achieve this goal by the binary box-based mask (box_mask_target) as following,

           mask_scores_phi = mask_scores_concat * box_mask_target
           img_target_wbox = img_target * box_mask_target
           high_feature = torch.cat((deep_stru_feature_img, deep_stru_feature_lst), dim=1) * box_mask_target

The region outside gt box for level set is zero, which are not participate in optimization.

You can remove the box_mask_target to run on the whole image.