search

Megvii-BaseDetection / DenseTeacher

DenseTeacher: Dense Pseudo-Label for Semi-supervised Object Detection
Apache License 2.0
120 stars 12 forks source link

question about position regression consistency #6

Closed JayYangSS closed 2 years ago

JayYangSS commented 2 years ago

I found that there is no difference data augmentation related to coordinate transformation in your code， student model and teacher model use the same coordinate transfrom, so the consistency constraint supervision of position prediction may not be enough. Have you done any relevant experiments?

Joker316701882 commented 2 years ago

@JayYangSS The inputs for the student model and the teacher model are different, they are still forced to produce the same outputs. This is the ''consistency constraint'' for the regression.

The geometrical augmentation may improve the performance, but we didn't try it. You can implement it by yourself and let us know if it works.

Joker316701882 commented 2 years ago

Since there is no further response from the questioner, this issue is closed for now. Reopen it when necessary.