search

httle / ARS-DETR

Apache License 2.0
42 stars 7 forks source link

The presence of NaN values #13

Open chengjc2019 opened 5 months ago

chengjc2019 commented 5 months ago

During my experiment, I modified the network structure of the bbox head, but the features extracted by the backbone contained NaN values, leading to errors in the assigner. This issue didn't occur if I didn't make modifications. I have two areas of confusion:

  1. Why does modifying the bbox head affect the extraction of backbone features, and why does it only affect certain images?
  2. I noticed in the source code of the rotated DETR model that the author debugged this issue in the forward_train function (outputting the image name if NaN is present). Have you encountered the same problem? Finally, how did you resolve it? Thank you.
httle commented 4 months ago

What specific modifications did you make? I'm sorry that I didn't modify head much, so I didn't encounter the problems you mentioned.