facebookresearch / Mask2Former

Code release for "Masked-attention Mask Transformer for Universal Image Segmentation"
MIT License
2.59k stars 388 forks source link

Very precise borderline. #195

Open sushilkhadkaanon opened 1 year ago

sushilkhadkaanon commented 1 year ago

I have trained the Mask2former model. The instance segmentation result has a very much precise borderline as compared to the results from MaskRCNN. I can speculate that this is because, during the feeding of feature maps from the pixel decoder to the transformer decoder, it adds positional embedding also which helps the model to learn to relate the same object from low to high resolution. Is it so I'm also not sure?