David Fan

The issue is with the way images are padded in the [RandomScaleCrop](https://github.com/jfzhang95/pytorch-deeplab-xception/blob/master/dataloaders/custom_transforms.py#L88) data augmentation, when the shorter side after rescaling is not large enough to be cropped. As is, the...

Could also try substituting in a sparse / linear self-attention mechanism

https://github.com/harritaylor/torchvggish/pull/19 Sending the model to GPU works fine but PyTorch will complain `RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same` unless the audio tensor is also...

I observe an increase in memory during the course of training when using the AudioReader during distributed PyTorch training. This doesn't happen when using the VideoReader

I just copied what Browser Genome did although it's true they might not have worried about efficiency

Uncommented out the .2bit download code and everything should work now! My copy is working at https://dfan97.github.io/ubit2/ with all the features (except the .2bit download).

The `lock_text_transformer` function is not correct because it doesn't add a case for handling a list of modules. It will raise a TypeError

There's a DOM 5 in Safari when the image source is encoded to base64 (line 247 in napster's pull request) because Safari doesn't like escape characters but Chrome is okay...

Added https://github.com/NYTimes/svg-crowbar/pull/25

Is there a reason why the checkpoint saving uses torch.save()? It seems that the full model weights are stored per rank instead of the sharded model weights, so the overall...