remove (un-)squashing logic from crop mask kernels

[pmeier]

After we have put the (un-)squashing logic in, it seems we have refactored the image kernel in a way so we no longer need the extra mask handling.

2d masks are tests as segmentation masks in

vision/test/test_transforms_v2_refactored.py

Lines 2637 to 2639 in f96deba

	@pytest.mark.parametrize("make_mask", [make_segmentation_mask, make_detection_mask])
	def test_kernel_mask(self, make_mask):
	check_kernel(F.crop_mask, make_mask(self.INPUT_SIZE), **self.MINIMAL_CROP_KWARGS)

and

vision/test/test_transforms_v2_refactored.py

Lines 3675 to 3677 in f96deba

	@pytest.mark.parametrize("make_mask", [make_segmentation_mask, make_detection_mask])
	def test_kernel_mask(self, make_mask):
	check_kernel(F.center_crop_mask, make_mask(), output_size=self.OUTPUT_SIZES[0])

cc @vfdev-5

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/8005

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 1 Unrelated Failure

As of commit 4a6f093 with merge base f96deba ():

NEW FAILURE - The following job has failed:

• Build M1 Wheels (gh)

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• unittests-macos (3.8, macos-m1-12) / macos-job (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[NicolasHug]

Touching on what we discussed offline: in main, we convert masks to 3D and we have correctness tests for (3D) images. So indirectly, we have correctness tests for masks.

With this PR, as-is, we're loosing any correctness test we have on masks. Yes we still have the kernel checks, but these don't test correctness against a reference.

[NicolasHug]

With this PR, as-is, we're loosing any correctness test we have on masks. Yes we still have the kernel checks, but these don't test correctness against a reference.

There is some minimal degree of correctness check, but it's very indirect: _check_kernel_batched_vs_unbatched, will ensure that the batched mask (3D, treated as an image, which are tested for correctness) are transformed the same as the unbatched ones (2D).

It only works for these 2 kernels because the C dim of images is treated as N dim of 2D masks.