add support for single channel images #174

edgarriba · 2017-05-24T00:28:07Z

No description provided.

fmassa

Hey, sorry for the delay in reviewing.
This looks good for the most part, I made some small comments (mostly cosmetics). Could you address them?

torchvision/transforms.py

@@ -43,7 +43,10 @@ class ToTensor(object):
    def __call__(self, pic):
        if isinstance(pic, np.ndarray):
            # handle numpy array
-            img = torch.from_numpy(pic.transpose((2, 0, 1)))
+            if len(pic.shape) >= 3:


torchvision/transforms.py

+            if len(pic.shape) >= 3:
+                img = torch.from_numpy(pic.transpose((2, 0, 1)))
+            else:
+                img = torch.from_numpy(pic.reshape((1,) + pic.shape))


torchvision/transforms.py

@@ -89,6 +92,8 @@ def __call__(self, pic):
        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:


torchvision/transforms.py

@@ -89,6 +92,8 @@ def __call__(self, pic):
        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:
+            npimg = np.reshape(npimg, npimg.shape + (1,))


alykhantejani · 2017-09-16T11:38:55Z

@edgarriba are you able to make the changes requested by @fmassa?
Also, it would be good to add another case to test_to_tensor to test for single channel images.

edgarriba · 2017-09-16T16:40:22Z

@alykhantejani @fmassa sure, I'll allocate some time during this weekend

test/test_transforms.py

@@ -1,6 +1,7 @@
 import torch
 import torchvision.transforms as transforms
 import unittest
+from parameterized import parameterized


torchvision/transforms.py


        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)


test/test_transforms.py

@@ -1,6 +1,7 @@
 import torch
 import torchvision.transforms as transforms
 import unittest
+from parameterized import parameterized


test/test_transforms.py

+        ('3channel', 3, 4, 4),
+        ('1channel', 1, 4, 4),
+    ])
+    def test_pil_to_tensor(self, _, channels, height, width):


alykhantejani · 2017-09-18T15:32:33Z

@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?

utkuozbulak · 2018-06-07T02:44:17Z

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

    
           if not(_is_pil_image(pic) or _is_numpy_image(pic)): 
        
               raise TypeError('pic should be PIL Image or ndarray. Got {}'.format(type(pic)))

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

    
           if isinstance(pic, np.ndarray): 
        
               # handle numpy array 
        
               img = torch.from_numpy(pic.transpose((2, 0, 1)))

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?

fmassa · 2018-06-07T13:33:01Z

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.
I'll open an issue with the description of the proposal once I get the chance.

utkuozbulak · 2018-06-08T04:30:04Z

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

fmassa requested changes Jun 15, 2017

View reviewed changes

fmassa added the awaiting response label Sep 16, 2017

edgarriba added 3 commits September 17, 2017 12:14

add support for single channel images

c7d017a

improve dimension statements and add separated test cases

1d4f85e

autopep8 style format

070940c

edgarriba commented Sep 17, 2017

View reviewed changes

test/test_transforms.py

@@ -1,6 +1,7 @@

import torch

import torchvision.transforms as transforms

import unittest

from parameterized import parameterized

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

alykhantejani suggested changes Sep 18, 2017

View reviewed changes

alykhantejani reviewed Sep 18, 2017

View reviewed changes

test/test_transforms.py

('3channel', 3, 4, 4),

('1channel', 1, 4, 4),

])

def test_pil_to_tensor(self, _, channels, height, width):

This comment was marked as off-topic.

Sign in to view

facebook-github-bot added the cla signed label Jan 20, 2021

rajveerb pushed a commit to rajveerb/vision that referenced this pull request Nov 30, 2023

adding tag to handle evaluations by iterations (pytorch#174)

d2b87dc

* adding tag to handle evaluations by iterations * better naming * better naming round 2 * update the python package version

edgarriba closed this Dec 5, 2023

add support for single channel images #174

add support for single channel images #174

Uh oh!

Conversation

edgarriba commented May 24, 2017

Uh oh!

fmassa left a comment

Choose a reason for hiding this comment

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

alykhantejani commented Sep 16, 2017

Uh oh!

edgarriba commented Sep 16, 2017

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

alykhantejani commented Sep 18, 2017

Uh oh!

utkuozbulak commented Jun 7, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fmassa commented Jun 7, 2018

Uh oh!

utkuozbulak commented Jun 8, 2018

Uh oh!

Uh oh!

utkuozbulak commented Jun 7, 2018 •

edited

Loading