The Vision Data Class#

The VisionData data class is the base class for all computer vision datasets, and represent a base CV task in deepchecks. It wraps PyTorch DataLoader together with model related metadata and contains additional data and general methods intended for easily accessing metadata relevant for validating a computer vision ML models.

For more info, please visit the API reference page: VisionData


The VisionData class represents a base CV task, and ignores the labels of the dataset, and the predictions of the model. It is mainly used for checks that doesn’t require them, such as ImagePropertyDrift

Class Properties#

The common properties are:

  • label_map - A dictionary mapping class ids to their names.

  • name - The dataset name to present in the displays instead of Train or Test.

Accepted Image Format#

All checks in deepchecks require images in the same format. They use the batch_to_images() function in order to get the images in the correct format.

The accepted format for a batch of images is an iterable of cv2 images. Each image in the iterable must be a [H, W, C] 3D numpy array. The first dimension must be the image y axis, the second being the image x axis, and the third being the number of channels. The numbers in the array should be in the range [0, 255]. Color images should be in RGB format and have 3 channels, while grayscale images should have 1 channel. The dtype of the array should be uint8.


from import VisionData

class NormalizedImagesData(VisionData):
    """Implement a VisionData class for PIL images."""

    def batch_to_images(self, batch):
        """Convert a batch of images to a list of PIL images.

        batch : torch.Tensor
            The batch of images to convert.

            A list of PIL images.

        # Assuming batch[0] is a batch of (N, C, H, W) images, we convert it to (N, H, W, C)/
        imgs = batch[0].detach().numpy().transpose((0, 2, 3, 1))

        # The images are normalized to [0, 1] range based on the mean and std of the ImageNet dataset, so we need to
        # convert them back to [0, 255] range.
        mean = [0.485, 0.456, 0.406]
        std = [0.229, 0.224, 0.225]
        imgs = std * imgs + mean
        imgs = np.clip(imgs, 0, 1)
        imgs *= 255

        return imgs

# Now, in order to test the class, we can create an instance of it:
data = NormalizedImagesData(your_dataloader)

# And validate the implementation: