About loss in Eq1

[JAYATEJAK]

Hi @ducminhkhoi , can you please let me know where exactly the loss is calculated in Equation 1?. I am checking through file in (iFSCRCNN-> modeling -> roi_heads -> fastrcnn.py), but I didn't get any variance matrix for classifier weights that mentioned in equation.

[ducminhkhoi]

Hi, it is the L473 and L479 of
https://github.com/ducminhkhoi/iFS-RCNN/blob/7b40243b5dd45e8512d12dc3963829b96eb4bb4f/iFS-RCNN/modeling/roi_heads/fast_rcnn.py#L473.
Basically, it is the standard sigmoid focal loss with the new sigmoid computation in Eq. (5) and KL div loss (i.e., minimize the variance).

Best,

[JAYATEJAK]

Thank you so much for clarification!. I have the doubt in the line 479 -> return_losses['loss_reg'] = self.box_predictor.var.mean() -> Where this self.box_predictor function is defined? (from where this function is imported)

Thanks in advance

[ducminhkhoi]

It is in https://github.com/ducminhkhoi/iFS-RCNN/blob/8a47ef216d6e344d0c7e560cb24fcd8b438bc929/iFS-RCNN/modeling/roi_heads/roi_heads.py#L569 and https://github.com/ducminhkhoi/iFS-RCNN/blob/8a47ef216d6e344d0c7e560cb24fcd8b438bc929/iFS-RCNN/modeling/roi_heads/roi_heads.py#L434

[JAYATEJAK]

Thank you so much.