Alain Riou
Alain Riou
Commenting the line ``` self.dataset = _OriginalDataset(self.safe_dataset) ``` in `nonechucks/dataloader.py` seems to solve the issue (in PyTorch 1.5.0) in the basic example I tried on. However I assume that it...
Hi, I did not solve it but found 2 ways to circumvent it: First one is to avoid using `torch.einsum`, which can be replaced by axis permutations and tensor multiplications....
Hi Ammon, Thanks a lot for sharing what you found out, I just read carefully the paper and it seems that you are absolutely right and that I misunderstood the...
Hi! Since this issue is quite annoying but very easy to fix, I'd be happy to submit a PR fixing it. Does it sound good to you? Thanks!
Oh actually I just realized that this import issue has already been fixed in the `main` branch, however when you install madmom with `pip` it is an older version of...
Maybe you can have a look at ffcv-ssl, a fork of this library that efficiently implements loaders for pairs of inputs, as well as more data augmentations