search

Megvii-BaseDetection / DenseTeacher

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

problem with prediction consistency #3

Closed eleanor11 closed 2 years ago

eleanor11 commented 2 years ago

Hi, I notice the input of teacher and student model are different (unsup_strong for student and unsup_weak for teacher). And there are some crop and flip operation in data augmentation. So will there be some problem with the prediction consistency when directly computing the loss with teacher's ans student's predictions (logits, deltas and quality) ? Or is there any extra transform operations to the predictions in this project?

ZRandomize commented 2 years ago

Please notice that the StrongAug is applied on the results of WeakAug, and the StrongAug it self contains no geometric transformation. In that case, the student and teacher model should be consistent

eleanor11 commented 2 years ago

okay，thank you！