YOLOv3-PyTorch CUDA out of memory!

CUDA out of memory!

Open JamesZhan107 opened this issue 3 years ago • 7 comments

When i train custom data , It always gives this error.

Are there any parameters I need to set ???

Oct 09 '20 06:10 JamesZhan107

I solved this problem after add --image-size in train command! But the mAP is very low...

Oct 09 '20 07:10 JamesZhan107

@Lornatang

Oct 10 '20 11:10 JamesZhan107

I don't seem to have encountered this problem, and if there is not enough batch, there may be a problem loading the dataset.

Oct 10 '20 12:10 Lornatang

I don't seem to have encountered this problem, and if there is not enough batch, there may be a problem loading the dataset.

The mAP is still very low after i modify the batch-size to 64.

But when i trained on ultralytics/yolov3 , it works well. @Lornatang

Oct 11 '20 02:10 JamesZhan107

when i use your mobilenetv3-large cfg file on ultralytics/yolov3, It produces the following error.

Is this a problem with the cfg file or what function needs to be modified？ @Lornatang

I'm sorry to disturb you !

Oct 11 '20 03:10 JamesZhan107

I didn't test enough on the mobilenetv3 large configuration file. Don't use it first

Oct 11 '20 04:10 Lornatang

OK, there is no problem with v1 and v2. Mobilenetv3 large and small configuration file has something wrong.

Oct 12 '20 01:10 JamesZhan107