Plt.imshow shows weird data augmentation when I have none

1

The below piece of code gives me back really weird data augmented image. There are no other transformations besides the Resize. I am using pytorch 0.40

Screenshot_57

transformed_dataset = data_sets(csv_path=csv_path,
                                image_path = image_path,
                                transform = transforms.Compose([
                                    transforms.Resize(256),
                                    transforms.ToTensor(),
                                    
                                ]) )

dataloader = DataLoader(transformed_dataset, batch_size=2,
                        shuffle=True, num_workers=0)

for idx, _ in enumerate(dataloader):
    #print(_['input'].size())
    for _ in _['input']:
        img = _.view(256, 256, 3).numpy()
        
        plt.imshow(img)

I am really not familiar with matplotlib so maybe this is the default behavior?

2

What is the photo supposed to look like? Several possibilities are the data is floating point and imshow is interpreting it as an integer (expected range is 0-255) but the data is float (expected range is 0-1) or vice versa.

3

Its suppose to be a single color image of a black and white line art. I didn’t want to convert it to gray scale because I want to train a classifier to ID the between color and black/white lineart.test_issue

5

Could you try to use img = img.permute(1, 2, 0) instead of the view call?

6

Thanks! That fixed it, do you know why view didn’t work? Edited: A quick loop shows permute works for some images but not on all of them?

Edited: Nvm there was left over old code. Thanks for the solution!