IA-SSD
IA-SSD copied to clipboard
center_loss_cls is too big
Hi, yifan, I tried to train the model on my own dataset, but I found the value of center_loss_cls is too big(from 100 to 300) when training. I want to know why and how to solve that problem.
Sorry I don't know much about your own data. These might be caused by inaccurate predictions of candidate points, so you can turn down the vote weight in LOSS_WEIGHTS, and adjust the EXTRA_WIDTH in TARGET_CONFIG. May it help you!