huggingface/pytorch-image-models · warning

range should be of kind (min, max)

Error message

range should be of kind (min, max)

What it means

timm's RandomResizedCrop (and INTERCEPT mode variants) warn when scale=(min,max) or ratio=(min,max) is passed with min > max. Unlike naflex's _validate_range, here the values are NOT swapped — the range is used as-is, so sampling can behave unexpectedly.

Source

Thrown at timm/data/transforms.py:193

        size: expected output size of each edge
        scale: range of size of the origin size cropped
        ratio: range of aspect ratio of the origin aspect ratio cropped
        interpolation: Default: PIL.Image.BILINEAR
    """

    def __init__(
            self,
            size,
            scale=(0.08, 1.0),
            ratio=(3. / 4., 4. / 3.),
            interpolation='bilinear',
    ):
        if isinstance(size, (list, tuple)):
            self.size = tuple(size)
        else:
            self.size = (size, size)
        if (scale[0] > scale[1]) or (ratio[0] > ratio[1]):
            warnings.warn("range should be of kind (min, max)")

        if interpolation == 'random':
            self.interpolation = _RANDOM_INTERPOLATION
        else:
            self.interpolation = str_to_interp_mode(interpolation)
        self.scale = scale
        self.ratio = ratio

    @staticmethod
    def get_params(img, scale, ratio):
        """Get parameters for ``crop`` for a random sized crop.

        Args:
            img (PIL Image): Image to be cropped.
            scale (tuple): range of size of the origin size cropped
            ratio (tuple): range of aspect ratio of the origin aspect ratio cropped

        Returns:

View on GitHub (pinned to 9a5261e31b)

Solutions

1. Rewrite ranges as (min, max)
2. Audit any augmentation config where ranges were authored as (max, min)

Example fix

# before
tfm = RandomResizedCrop(224, scale=(1.0, 0.08), ratio=(4./3., 3./4.))
# after
tfm = RandomResizedCrop(224, scale=(0.08, 1.0), ratio=(3./4., 4./3.))

assert scale[0] <= scale[1] and ratio[0] <= ratio[1], 'ranges must be (min, max)'

When it happens

Trigger: Creating RandomResizedCrop(224, scale=(0.9, 0.5)) or ratio=(4/3, 3/4); porting configs written as (max, min).

Common situations: Copy-pasted torchvision examples with reversed ratio; hand-tuned augmentation configs. Because no swap happens, downstream torch.empty(...).uniform_(lo, hi) with lo>hi raises or yields empty samples — fix the order.

Related errors

• {name.capitalize()} range reversed. Swapping.
• Input image must have positive dimensions, got H={height}, W
• Transform returned None for index {idx}. Skipping sample.
• final_scale_range values should ideally be between 0.0 and 1
• Final scale randomization ({scale_factor:.2f}) resulted in s

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/c01da15412dc3906. Report an issue: GitHub.