cls loss is increasing

[jacksonsc007]

Hi, @hyz-xmaster I noticed that the cls_loss using varifocal loss does not decrease until roughly 1000 iterations. Could you give us some insights?

[hyz-xmaster]

You may try a lower learning rate or more powerful features like using DCN to extracting features to predict IoU-aware cls scores.

[jacksonsc007]

@hyz-xmaster Thanks for your prompt reply. I checked your training log and found that the cls_loss kept loitering around for 700 iterations, is this normal or a problem we need to address?

[hyz-xmaster]

@jacksonsc007 Well, there is a warm-up stage in training where the learning rate increases linearly from a very small value to the initial learning rate. In general, this process takes 500 iterations. See here . During this stage, the losses move up and down. So it's normal.

[jacksonsc007]

@hyz-xmaster Thank you.