search

JosephKJ / OWOD

(CVPR 2021 Oral) Open World Object Detection
https://josephkj.in
Apache License 2.0
1.04k stars 155 forks source link

Why use the proposal of an unknown class to regress the gt box of a known class #44

Closed fan945327815 closed 3 years ago

fan945327815 commented 3 years ago

Thanks for your greate job! I don't know if I understand correctly. In roi_head stage, is it reasonable to use the proposal of the unknown class to regress the gt box of the known class with the lowest IOU

dyabel commented 3 years ago

Thanks for your greate job! I don't know if I understand correctly. In roi_head stage, is it reasonable to use the proposal of the unknown class to regress the gt box of the known class with the lowest IOU

I am also confused about this

JosephKJ commented 3 years ago

The proposals that are auto-labelled as unknowns are used only in the contrastive clustering. It is used to learn the classification/regression head. Does this resolve the confusion?

Please re-open this issue for more discussion.

TabGuigui commented 1 year ago

hi there, The proposals that are auto-labelled as unknowns are still used to train the regression head in the code? Is there something improper that the unknown proposals have a corresponding gt boxes to regress?