Why are only 2 of the RT-DETR v2 implemented losses actually used?

šŸ¤—Transformers
1

Hi,

Iā€™ve been studying the RTDetrLoss implementation in the transformers library and noticed something interesting. The class implements several loss functions:

  • loss_labels_vfl
  • loss_labels (cross-entropy)
  • loss_cardinality
  • loss_boxes
  • loss_masks
  • loss_labels_bce
  • loss_labels_focal

However, looking at the code, it seems only ā€œvflā€ and ā€œboxesā€ losses are actually used in practice (these are the only ones included in self.losses = ["vfl", "boxes"]).

Iā€™m curious about why the other loss functions are implemented but not utilized. Is this consistent with the RT-DETR paperā€™s implementation? Or perhaps were the other losses tested empirically but found to be less effective?

Has anyone experimented with enabling the other loss functions like the focal loss or BCE loss? Iā€™d appreciate any insights about the design choices here.

Hope you guys can help! Thank you.

1 Like
2

If anyone has an answer it would be very helpful.

1 Like
3

I did some quick research at the time, but couldnā€™t find any useful information. It seems that those parameters already existed in RT-DETR (not v2), but I couldnā€™t figure out why they were implemented.

1 Like
4

ok thank you

1 Like