batch_size error

Open boringwar opened this issue 4 years ago • 1 comments

Hi, when conducting the classification experiment on imagenet, the default batch_size is 32 x 8 on my machine if I follow the instructions. In your training log, the batch_size is 32 x 64. Why is this difference?

Version: latest mmcv,mmcls, mmdet as well as pytorch 1.7.1, torchvision 0.8.2.

Apr 07 '21 02:04 boringwar

I noticed that your experiments is done on 64 gpu cards, I'm wondering how to achieve the same performance with only 8 cards.

Apr 07 '21 05:04 boringwar