Hi, Thanks for the great work! When I tried to use the Deformable-DETR on my own dataset, I found that the PostProcess() in models/deformable_detr.py always selected the top-100 predictions as the final results for an input image. Because of using the focal loss, there was no non-object class in Deformable-DETR, so the 100 results were all the objects. Did I misunderstand something?
However, only several objects per image on my dataset, so the 100 predicted objects would lead to a very low 'precision'. How could I select the 'real' objects from all the outputs? Maybe set a confidence threshold based on the score?
brent88310070
commented
4 months ago
Hi, Thanks for the great work! When I tried to use the Deformable-DETR on my own dataset, I found that the
PostProcess()inmodels/deformable_detr.pyalways selected the top-100 predictions as the final results for an input image. Because of using the focal loss, there was nonon-objectclass in Deformable-DETR, so the 100 results were all theobjects. Did I misunderstand something? However, only several objects per image on my dataset, so the 100 predicted objects would lead to a very low 'precision'. How could I select the 'real' objects from all the outputs? Maybe set a confidence threshold based on the score?