Updated on June 10, 2022 at 09:54:00

—————————————————————————————————

The problem is that the number of categories is not modified

————————————————————————————————————

This issue has been reported on github: https://github.com/open-mmlab/mmdetection/issues/7298

Problem Description:

I used 2 2080ti cards for training, and after a few iterations in epoch1, it stopped moving, but the gpu utilization was still floating; then I used one card for training, and the problem under this issue was reported.

Solution:

I checked all my pictures, each picture has a frame, the frame position is normal, and then I use the error message to try the method find_unused_parameters=True (add it under if distributed in mmdet/apis/train.py), under two cards Currently training 2 epochs without error, continue to observe