AttributeError: module 'transforms' has no attribute 'Normalize'

vision
1

Hi. I’m using torchvision 0.13.0.

I had this problem:

TypeError: Expected input images to be of floating type (in range [0, 1]), but found type torch.uint8 instead

When I was attempting to do this:

import transforms as T


def get_transform(train):
    transform = []
    # converts the image, a PIL image, into a PyTorch Tensor
    transform.append([T.PILToTensor(), T.Normalize()])
    if train:
        # during training, randomly flip the training images
        # and ground-truth for data augmentation
        transform.append(T.RandomHorizontalFlip(0.5))
    return T.Compose(transform)


model = torchvision.models.detection.fasterrcnn_resnet50_fpn(weights=True)
dataset = four_chs(root = '/home/jn/Downloads/', transforms = get_transform(train=True))   

data_loader = torch.utils.data.DataLoader(
 dataset, batch_size=1, shuffle=True, num_workers=2,
 collate_fn=utils.collate_fn)

# For Training
images,targets = next(iter(data_loader))
images = list(image for image in images)
targets = [{k: v for k, v in t.items()} for t in targets]
output = model(images,targets)   # Returns losses and detections

I searched some and found something that suggested using a Normalize transform. I’m not sure how to string together multiple transforms, so I tried the code above. That returns this error:

AttributeError: module ‘transforms’ has no attribute ‘Normalize’

The Normalize transform seems to be defined, so I’m not sure why this error happens. How do I address this?

2

It seems you are importing a custom transforms module and not torchvision.transforms, which doesn’t seem to have the Normalize transformation.
Could you check where this module is defined and also check why it’s used?

3

Hi ptrblck. I used this section earlier in the code. Sorry I didn’t include it in first post.

if os.path.isdir("vision") == True:
    print("vision present")
else:
    Repo.clone_from("https://github.com/pytorch/vision.git", "vision")

shutil.copy('vision/references/detection/utils.py', 'utils.py')
shutil.copy('vision/references/detection/transforms.py', 'transforms.py')
shutil.copy('vision/references/detection/coco_eval.py', 'coco_eval.py')
shutil.copy('vision/references/detection/engine.py', 'engine.py')
shutil.copy('vision/references/detection/coco_utils.py', 'coco_utils.py')

I used that because it’s mostly the same as what’s provided here:

That Colab tutorial relies on torchvision 0.12.0, I think, and I’m trying to use 0.13.0 at this point, so I’m using most of the same code.

I see “vision/torchvision/transforms at main · pytorch/vision · GitHub”. I should use that instead? It seems to have the transforms module, but not engine. I suppose I should copy both repos and import the requisite modules accordingly.

4

Thanks for the explanation.
Import the torchvision.transforms and your local transforms as different modules and it should work.
E.g.

from transforms as det_transforms
import torchvision.transforms as transforms

and use these two names to separate the transformations.

1 Like