Mmdetection: Can you explain more about how to distributed train on two machines each with 4 cores?

[xbjxh]

Thank you so much! One machine works fine too me. But I am not sure about the setting in the configure and how to train it on two machine.

[hellock]

We just use torch.distributed.launch to start the training processes, see tools/dist_train.sh for details. If you want to train on two machines, you need to follow the "Multi-Node multi-process distributed training" part in the documentation.

Just a reminder, If multiple nodes are not connected with high-speed hardwares, it will be slow.