DeLightCMU / CASD

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The memory of the code

AiLearningCode opened this issue · comments

Hi,Nice work!

Thanks for releasing the code, I have a question about the GPU memory of this code. I ran the default parameter to train the network, like bs=1, net=vgg16. I have seen that the required GPU memory was 24G,When I used four 1080ti to train the network, I met the out of memory error, the memory of 1080ti is 11G. When are you going to release code of mutil gpu training ?

Thank you very much!

Hi,
If you use 1080ti, please see here:vasgaowei/pytorch_MELM#20.
Based on other wsod repositories, it looks like 8 GPU training is much better than single GPU training. So I hope I can implement everything in MMdetection around Christmas.