search

Megvii-BaseDetection / DenseTeacher

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

Could you summarize the arch differences from the original FCOS model? #19

Closed floveqq closed 1 year ago

floveqq commented 1 year ago

about the arch like loss,bn,dataaug and so on? thanks

ZRandomize commented 1 year ago

for supervised loss, cls loss and reg loss are same as in FCOS, using FocalLoss and IoU loss. The target of quality branch is replaced by IoU rather than centerness, using same BCE loss; quality branch moved to reg subnet. These mdifications following this work. We do not change bn. DataAug is not a part of model arch, we use augs same as in Unbiased Teacher.