add support for single channel images #174
Conversation
torchvision/transforms.py
Outdated
| if isinstance(pic, np.ndarray): | ||
| # handle numpy array | ||
| img = torch.from_numpy(pic.transpose((2, 0, 1))) | ||
| if len(pic.shape) >= 3: |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
torchvision/transforms.py
Outdated
| if len(pic.shape) >= 3: | ||
| img = torch.from_numpy(pic.transpose((2, 0, 1))) | ||
| else: | ||
| img = torch.from_numpy(pic.reshape((1,) + pic.shape)) |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| if torch.is_tensor(pic): | ||
| npimg = np.transpose(pic.numpy(), (1, 2, 0)) | ||
| assert isinstance(npimg, np.ndarray), 'pic should be Tensor or ndarray' | ||
| if len(npimg.shape) < 3: |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
torchvision/transforms.py
Outdated
| npimg = np.transpose(pic.numpy(), (1, 2, 0)) | ||
| assert isinstance(npimg, np.ndarray), 'pic should be Tensor or ndarray' | ||
| if len(npimg.shape) < 3: | ||
| npimg = np.reshape(npimg, npimg.shape + (1,)) |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
@edgarriba are you able to make the changes requested by @fmassa? |
|
@alykhantejani @fmassa sure, I'll allocate some time during this weekend |
| import torch | ||
| import torchvision.transforms as transforms | ||
| import unittest | ||
| from parameterized import parameterized |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
|
||
| if accimage is not None and isinstance(pic, accimage.Image): | ||
| nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.float32) | ||
| nppic=np.zeros([pic.channels, pic.height, pic.width], dtype=np.float32) |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| import torch | ||
| import torchvision.transforms as transforms | ||
| import unittest | ||
| from parameterized import parameterized |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| ('3channel', 3, 4, 4), | ||
| ('1channel', 1, 4, 4), | ||
| ]) | ||
| def test_pil_to_tensor(self, _, channels, height, width): |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
@edgarriba Thanks for this - in general it looks good just a few small things to fix. I don't think we want to add another dependency to the library if we can avoid it. Would you be able to make this change and the linting ones? |
|
Checks for image/numpy channels are implemented with refactoring of transforms to functional and transforms in #311 vision/torchvision/transforms/functional.py Lines 43 to 44 in 50b2f91 Although single channel images (HxW) still produce following error: ValueError: axes don't match array due to vision/torchvision/transforms/functional.py Lines 46 to 48 in 50b2f91 Wouldn't it be easier to just expand these arrays with a simple if, like: ...
if isinstance(pic, np.ndarray): ## Existing
# handle numpy array
if pic.ndim == 2: ## Addition
# expand by one dimension as the last channel ## Addition
img = torch.from_numpy(pic.transpose((2, 0, 1))) ## Existing
...Either way, this PR might as well be closed as a clean up due to massive conflicts? |
|
I'm planning on modifying the underlying data structure that we use internally (instead of using a PIL image, have a custom class), so that those corner cases can be fixed. |
|
Sounds like a plan, I would like to work on that if you will need a pair of hands. |
* adding tag to handle evaluations by iterations * better naming * better naming round 2 * update the python package version
5 participants
No description provided.