tensor2image and dataloder length

[wangjiawen2013]

Hi,
I have two questions now

1. In pytorch, we can convert tensors to images and save on disk using from torchvision.utils import save_image. How to convert a tensor to image and save it on disk using burn ?
2. In pytorch, we can get batch numbers using len(dataloader). How to get the batch numbers in a dataloader in burn ? It seems that dataloader.num_items only returns the total number of samples, not the batch number.

[laggui]

1. There is no utility similar to the pytorch vision extension for save_image in Burn atm. But this should be fairly straightforward using the tensor data and constructing an image buffer to save (e.g., Rgb32FImage)
2. The DataLoader trait doesn't expose the number of batches in a dataloader. It returns an iterator to iterate over the batches any fixed-size assumption. The dataloader length can only be known for a fixed-size dataset, so even pytorch makes assumption for different dataset types and might report inaccurate length (as in their disclaimer). So for now we chose not to make any assumptions 🙂