Loss:nan When training Hico_det dataset with detr-r50-hicodet.pth

[user-lele]

Hello!
I have followed your README file to set up the environment and prepare the dataset. The "inference.py" script has also run successfully, producing the expected results. However, when running the benchmark dataset, I always get a loss of NaN. Lowering lr_head or follow the previous issues did not solve the issue. How should I resolve it?
My training command is
"DETR=base python main.py --pretrained checkpoints/detr-r50-hicodet.pth --output-dir outputs/pvic-detr-r50-hicodet"

[fredzzhang]

Hi @user-lele,

Could you refer to post #65 to see if it solves the problem?

Fred.