Jack Lindsey

O2 doesn't seem to resolve it -- for now, the simplest workaround in my particular case is probably to just use a smaller ResNet so I can increase the batch...

Got it, thanks for the info! One more question, if you don't mind -- did you try the "bio plausible" version on ImageNet as well?

Hi! Just wanted to second this, it would be great to have the code and a pre-trained model for the SupContrast network on ImageNet.

Ah I see, so there is no ImageNet support yet. Would you expect plugging in the ImageNet dataset to the current code to work? Thanks for the help!

Thanks a bunch! I had in mind the case of layers in which the number of inputs per node remains constant as the width goes to infinity -- I'm not...

Trying with one GPU gives me a different error: Traceback (most recent call last): File "train.py", line 94, in trainer.fit(model, train_loader) File "/home/jwl2182/.local/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py", line918, in fit self.single_gpu_train(model) File "/home/jwl2182/.local/lib/python3.7/site-packages/pytorch_lightning/trainer/distrib_parts.py", line...

Thanks! That fixed the second problem. I now get a new error -- see below -- running on one GPU (and still get the original error running on multiple GPUs)...

Thanks a lot for being so responsive, that fixed the issue and I am now able to train on one GPU. I still get errors (now a different error) with...

Thanks, just figured out the same fix. I'm sorry to keep bothering you, but unfortunately all this has brought me back to the original error: Traceback (most recent call last):...

To clarify further -- the glimpses are taken at three separate spatial scales ("zoom" levels), with radii separated by factors of 2. Max radius refers to the largest of these.