search

facebookresearch / unbiased-teacher-v2

PyTorch code for CVPR 2022 paper Unbiased Teacher v2 Semi-supervised Object Detection for Anchor-free and Anchor-based Detectors
MIT License
90 stars 19 forks source link

some questions about your article #10

Open Tourbillon opened 1 year ago

Tourbillon commented 1 year ago

I have read your article carefully, which is very impressive, but I also have some questions to ask you

  1. The Appedix Figure 1.(d)，where is the iou result from， from test set or pseudo result?
  2. Is the Listen2Student mechanism used in FCOS?
  3. The result of FCOS used in Unbiased Teacher V1 is not consistent with those in Dense Teachers. There is a big difference between the two results.
CoderZhangYx commented 1 year ago

hello, did you find out the answer of question3? During reading i as well found that fcos experiment result is not consistent with other paper! For example the fcos on supervised 10% coco results in 26map in other paper but only 20map in this paper. When i ran my own process, i found that i will get a map over 20 too.