How to use this code in pytorch1.0

[mingrui-xie]

Because the inplace_abn(the third party libs that this project used) have some bugs in 8 GPUs pytorch0.4, if we want to train in 8GPUs, we should update the newest bn in https://github.com/mapillary/inplace_abn.
But the newset bn now requires to use DistributedDataParallel instead of DataParallel. so, could you
please create a branch that use the newest bn in pytorch 1.0 ? or give me some advice that how to change this project to make it compatible with the newest bn.
Thank you very much!

[speedinghzl]

Good suggestion. But I'm afraid that I don't have the time to do this.
You can simply replace current inplace-ABN with the newest one. And add torch.distributed into train.py. Maybe you can get more information in Pytorch doc or some examples.

[Leodora]

I have tried the newest Inplace-abn in my model(pspnet), and you can follow the script in this page : https://oldpan.me/archives/pytorch-to-use-multiple-gpus . Meanwhile, make sure that you have followed the steps in Pytorch doc.