Closed Virgilzz closed 2 years ago
Hi @Virgilzz, thanks for your interest in our work.
It seems the error comes from the SyncBN layer. We convert all batch norm layers to SyncBN here, which only supports DistributedDataParallel (DDP) environment. You may need to comment out this line if you want to train the model on a single GPU or without DDP.
It has been solved. Thank you!
Hi,I've modified the settings for single or multiple GPUs
norm_cfg = dict(type='BN', requires_grad=True)
But there were still such errors. ![Uploading image.png…]() Does that mean I need to modify the training section in the mmseg source?