search

ChenhongyiYang / PGD

[ECCV 2022] Prediction-Guided Distillation for Dense Object Detection
MIT License
63 stars 12 forks source link

PGD for coupled head? #10

Open aidevmin opened 10 months ago

aidevmin commented 10 months ago

Did you try any experiments of PGD with coupled head? In this case we don't seperated classification and regression head. I want to apply your method to my model with coupled head. In this case I need to modify loss function, mask. Why do you spatial attention mask for classification task and not for regression task?

ChenhongyiYang commented 10 months ago

Hi, thank you for being interested in our paper. We do not really need a spatial attention mask for regression because the regression loss is only applied to foreground areas. For coupled heads, I think you may just try distilling your model following the classification distillation strategy.

aidevmin commented 10 months ago

@ChenhongyiYang Thank you for suggestion. I aslo thought use kd loss as the classification strategy.