search

linxid / Focus-DETR-mindspore

[ICCV 2023] Official implementation of the paper "Less is More: Focus Attention for Efficient DETR"
Apache License 2.0
75 stars 2 forks source link

What is the ground truth for training? #1

Open rginjapan opened 12 months ago

rginjapan commented 12 months ago

Thanks for your work. When training the focal loss and multi-category score predictor, do you need the foreground and background label (ground truth)

linxid commented 12 months ago

We use COCO label to generate foreground and background label. Pytorch code is avaliable in https://github.com/IDEA-Research/detrex/blob/main/projects/focus_detr/modeling/two_stage_criterion.py

rginjapan commented 12 months ago

Thanks for your reply, about multi-category score predictor, does it mean select the top-k score in foreground token? "multi-category" is a little bit confused me.

linxid commented 11 months ago

"Multi-category score predictor " is similar to topk token selector in the Encoder of DINO.