Closed JoyHuYY1412 closed 2 years ago
Thanks for your interest.
1). If the single GPU has enough memory for the mixed regions (which are cut from the other GPUs' unlabelled data), then you can perform all_gather for the inter-communication in the cluster. But, the speed will be dramatically slowed down I guess.
2). If it cannot handle the mixed region, then I have no idea for this situation.
Regards, Yuyuan
Thanks for your interest.
1). If the single GPU has enough memory for the mixed regions (which are cut from the other GPUs' unlabelled data), then you can perform all_gather for the inter-communication in the cluster. But, the speed will be dramatically slowed down I guess.
2). If it cannot handle the mixed region, then I have no idea for this situation.
Regards, Yuyuan
Thank you. According to your training script, you used two gpus with 4 labeled and 4 unlabeled each gpu. Did I understood right?
Hi Joy,
You can find my training details through this link for voc training details or this one for cityscapes training details.
You can click the experiment name and find out hardware details in the overview and hardware tags (e.g., in here).
Regards, Yuyuan
Hi, thank you for your work. I have a question about the CuTMIx.
Since the batch size for each GPU is 1 when using 8GPUs (conventional setting), I wonder how to perform CutMix when only one unlabeled image is in a GPU? Can you give me some advice?