qinenergy/cotta

Running Code with DDP

Closed this issue · 1 comments

Hello, I have another inquiry is it possible to run the of semantic segmentation code with DistributedDataParallel option ?

https://github.com/NVlabs/SegFormer/blob/639caf64aad8241c670c3a6a820ab0a9bde02b77/tools/test.py#L145-L151

I think it should be possible if you modify the function multi_gpu_test accordingly and use the DistributedDataParallel wrapper from above.

However, we didn't test it and will probably not invest time on it. This is because CoTTA is an inference time adaptation method and inference on multiple gpus may not be realistic in real world applications.