search

aim-uofa / AdelaiDet

AdelaiDet is an open source toolbox for multiple instance-level detection and recognition tasks.
https://git.io/AdelaiDet
Other
3.38k stars 650 forks source link

Is there a handy way to reproduce the extension experiment results in the BoxInst paper? #450

Open njuxx opened 3 years ago

njuxx commented 3 years ago

BoxInst is a really nice job! In your paper, you did an extension experiment which is a semi-supervised instance segmentation task. I wonder if there is a handy way to reproduce those results shown in your paper? Thank you.

tianzhi0549 commented 3 years ago

Just try to apply the BoxInst losses when the mask annotations are not available. The BoxInst losses can be found here https://github.com/aim-uofa/AdelaiDet/blob/7bf9d871909810e5a619eb1a59a5cd5aadf0e1c8/adet/modeling/condinst/dynamic_mask_head.py#L223

njuxx commented 3 years ago

Just try to apply the BoxInst losses when the mask annotations are not available. The BoxInst losses can be found here

https://github.com/aim-uofa/AdelaiDet/blob/7bf9d871909810e5a619eb1a59a5cd5aadf0e1c8/adet/modeling/condinst/dynamic_mask_head.py#L223

Thank you