SamsungLabs / tr3d

[ICIP2023] TR3D: Towards Real-Time Indoor 3D Object Detection
Other
141 stars 8 forks source link

The training performance result issue #33

Open linQian99 opened 1 month ago

linQian99 commented 1 month ago

I hope this message finds you well. Thank you for your great work. May I ask why changing the score_thr in config file will affect the performance result? Thank you for your time. image

This is 0.30 threshold image

This is 0.01 threshold image

filaPro commented 1 month ago

Hi @linQian99 , It's just the feature of mAP metric. It doesn't decrease if you predict more objects with very low confidences. It is also correct for 2d detection.

linQian99 commented 1 month ago

Thank you for your reply, but if the confidence threshold set to 0.01, the visualization result will shows a lot of bounding boxes, is that normal?

filaPro commented 1 month ago

Yes, for visualization something like 0.3 should be fine.