Pytorch

Recently, while working on a video dataset, I noticed that some of the videos contained a few corrupted frames. While running the training, my dataloader used to return an incorrect shaped tensor since it was not able to read the corrupted frame.