Ease the visualization of latent directions

[woctezuma]

There are two changes:

• a CLI argument to be more generic, which lets us sample more than 3 points, compared to:

stylegan2-pytorch/apply_factor.py

Lines 61 to 83 in 7be6e64

	direction = args.degree * eigvec[:, args.index].unsqueeze(0)
	img, _ = g(
	[latent],
	truncation=args.truncation,
	truncation_latent=trunc,
	input_is_latent=True,
	)
	img1, _ = g(
	[latent + direction],
	truncation=args.truncation,
	truncation_latent=trunc,
	input_is_latent=True,
	)
	img2, _ = g(
	[latent - direction],
	truncation=args.truncation,
	truncation_latent=trunc,
	input_is_latent=True,
	)
	grid = utils.save_image(
	torch.cat([img1, img, img2], 0),

• a transposition of the grid of results as I think it makes more sense to show one sample per row (instead of per column).

This is what the transposed results look like:

NB: I have submitted a bunch of PR. I have merged all of them in one of my branch for my personal use:
https://github.com/woctezuma/stylegan2-pytorch/tree/quality-of-life-merge
In case you want to merge most of them, this would make it easier for you. Just let me know and I would submit a PR.