During training, the loss is nan

[luoyq6]

python train.py --dataset coco_base --flip --net hanmcl --lr 0.001 --lr_decay_step 12 --bs 4 --epochs 12 --disp_interval 20 --save_dir models/hanmcl --way 2 --shot 3
I used this parameter for training and this phenomenon occurred.
iter 0/34468：rpn_cls: 2921850624.0000, rpn_box: 962897664.0000, rcnn_cls: nan, rcnn_box 102172360.0000, metric nan
iter 20/34468：rpn_cls: nan, rpn_box: nan, rcnn_cls: nan, rcnn_box nan, metric nan

[giangdip2410]

Same with me. I faced nan loss after 1,2 epoch. @infinity7428 : could you please help to check the issue ?

[winnie-nqw]

Same with me. I faced nan loss after 1,2 epoch. @infinity7428 : could you please help to check the issue ?

[yuan738]

Did you solve the problem please? I'm having the same problem and need help, thanks.

[infinity7428]

Probably, the code will be updated this week

[yuan738]

Thank you for your reply and hope to see the updated code soon.

[infinity7428]

I updated the code. If the same problem occurs, please give me a mention.

[yuan738]

Thank you!

[Nepenthes-zlc]

I still meet the problem. The rcnn_cls and metrics is nan