Making Vflip and Hflip in Tensor format #1465
Conversation
test/test_functional_tensor.py
Outdated
| vflipped_img = F_t.vflip(img_tensor) | ||
| vflipped_img_again = F_t.vflip(vflipped_img) | ||
|
|
||
| assert vflipped_img.shape == img_tensor.shape |
There was a problem hiding this comment.
Can we use self.assertEqual here? I'd prefer to move away from the raw asserts and use self.assert* methods from unittest for newer test
test/test_functional_tensor.py
Outdated
| vflipped_img_again = F_t.vflip(vflipped_img) | ||
|
|
||
| assert vflipped_img.shape == img_tensor.shape | ||
| assert torch.equal(img_tensor, vflipped_img_again) |
There was a problem hiding this comment.
I think I'd prefer using self.assertTrue here.
| """Vertically flip the given the Image Tensor. | ||
|
|
||
| Args: | ||
| img_tensor (Tensor): Image Tensor to be flipped. |
There was a problem hiding this comment.
Can you specify in the documentation what is the expected format of the tensor? Is it HxWxC or CxHxW?
| if not F._is_tensor_image(img_tensor): | ||
| raise TypeError('tensor is not a torch image.') | ||
|
|
||
| return img_tensor.flip(1) |
There was a problem hiding this comment.
Can you make the offset relative to the end of the dimension?
This way, we could make it also work for batches of images / videos.
So it would be something like
return img_tensor.flip(-2)| if not F._is_tensor_image(img_tensor): | ||
| raise TypeError('tensor is not a torch image.') | ||
|
|
||
| return img_tensor.flip(2) |
There was a problem hiding this comment.
Same comment, can you do .flip(-1) here?
|
@fmassa , got'em. Making the updates. |
Codecov Report
@@ Coverage Diff @@
## master #1465 +/- ##
==========================================
- Coverage 64.08% 64.07% -0.01%
==========================================
Files 80 81 +1
Lines 6328 6338 +10
Branches 973 975 +2
==========================================
+ Hits 4055 4061 +6
- Misses 1986 1988 +2
- Partials 287 289 +2
Continue to review full report at Codecov.
|
fmassa
left a comment
fmassa
left a comment
There was a problem hiding this comment.
Lint is still failing, and I have a minor nit but otherwise looks good to be merged, thanks!
I need to figure out what's happening with CircleCI though
|
|
||
|
|
||
| if __name__ == '__main__': | ||
| unittest.main() No newline at end of file |
There was a problem hiding this comment.
linter is failing due to the missing newline here
| """Vertically flip the given the Image Tensor. | ||
|
|
||
| Args: | ||
| img_tensor (Tensor): Image Tensor to be flipped in the form CXHXW. |
There was a problem hiding this comment.
nit: can you make it CxHxW or [C, H, W]? I think it's more readable.
|
@surgan12 the CircleCI error is very weird, it looks like it's trying to run from your account. Can you try creating a new branch in your fork of torchvision (with the current commits), and then creating a new PR? and then try again? |
|
Sure will do that ! |
3 participants
cc: @fmassa
Ref: #1375