search

fudan-zvg / SETR

[CVPR 2021] Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers
MIT License
1.05k stars 150 forks source link

Auxiliary loss #20

Closed yangshushuaige closed 3 years ago

yangshushuaige commented 3 years ago

Thank you for your great work. Auxiliary segmentation loss helps the model training and may lead to significant performance improvement. Can you provide some related results w/o auxiliary loss?

yangshushuaige commented 3 years ago

For SETR-Naive and SETR-PUP only using the final output of transformer, do you also use the Auxiliary segmentation loss to supervise intermediate layers of transformer?

lzrobots commented 3 years ago

Thank you for your great work. Auxiliary segmentation loss helps the model training and may lead to significant performance improvement. Can you provide some related results w/o auxiliary loss?

Thank you. We didn't report the results w/o auxiliary loss due to the space issue and you can have a try without the auxiliary loss.

For SETR-Naive and SETR-PUP only using the final output of transformer, do you also use the Auxiliary segmentation loss to supervise intermediate layers of transformer?

Yes. Please take a look of the config file pup and naive.