param `nrow` in `make_grid` is misleading

[alykhantejani]

import torch
import torchvision
label_img = torch.rand(100, 3, 32, 32)
x = torchvision.utils.make_grid(label_img, nrow=20, padding=10)
torchvision.transforms.ToPILImage()(x).save("example.png")

produces this image:

which has 5 rows and 20 columns. Although the docs say

nrow (int, optional) – Number of rows in grid. Final grid size is (B / nrow, nrow). Default is 8.

Which says the output will actually be (B/nrow, nrow). Perhaps this param should be called ncol, or we should change the output shape to be (nrow, B/nrow)?

I think, to keep backward compatibility changing the output shape to (nrow, B/nrow) would make more sense.

Happy to send a PR if you agree?

[fmassa]

Maybe we could change the docs to read "number of images in a row"? This way we don't change the API, and it still makes sense from the parameter name?

[alykhantejani]

Yeah - this is a lot easier and wont have "unexpected" changes for people who are used to using it in its current form. I'll send a PR for this.

[fmassa]

Fixed via #250