setting max_num_instances, max_detections_per_image

[Ekta246]

I still have this question, and I cannot find a perfect answer for the setting of max_num_instances and max_detections_per_image in the anchors.py

However, for now since I am working only on 1 class and 1 bounding box(only 1 GT label) per image I find the task easier with setting both the max points to 1.
I see my model now gives bbox with scores 98% and higher. I am confused if this going in correct direction?Or the max points should be increased(set higher than 1)?

I would really appreciate the answer to this question

TIA

[Ekta246]

I still have this question, and I cannot find a perfect answer for the setting of max_num_instances and max_detections_per_image in the anchors.py

However, for now since I am working only on 1 class and 1 bounding box(only 1 GT label) per image I find the task easier with setting both the max points to 1.
I see my model now gives bbox with scores 98% and higher. I am confused if this going in correct direction?Or the max points should be increased(set higher than 1)?

I would really appreciate the answer to this question

TIA

@MichaelMonashev I think you had asked the same question

[MichaelMonashev]

I just patched the code.

[jaideep11061982]

how one can get class wise confidence rather directly idx currently in output['detections][:,:,:,5]?
@MichaelMonashev @rwightman any suggestions.