PaddlePaddle/PaddleOCR · error · ValueError

'size' must be a list or tuple of two numbers, but got {size

Error message

'size' must be a list or tuple of two numbers, but got {size}

What it means

When the IAA augmenter type is 'Resize', PaddleOCR maps the imgaug-style 'size' argument to its internal scale_range format. The mapping requires 'size' to be a list or tuple of exactly two numbers; anything else (scalar, string, 3-element list) raises ValueError with the offending value embedded in the message.

Source

Thrown at ppocr/data/imaug/iaa_augment.py:140

                cls = getattr(A, augmenter_type_mapped)
                return cls(
                    **{
                        k: self.to_tuple_if_list(v)
                        for k, v in augmenter_args_mapped.items()
                    }
                )
        else:
            raise RuntimeError("Unknown augmenter arg: " + str(args))

    # Map arguments to expected format for each augmenter type
    def map_arguments(self, augmenter_type, augmenter_args):
        augmenter_args = augmenter_args.copy()  # Avoid modifying the original arguments
        if augmenter_type == "Resize":
            # Ensure size is a valid 2-element list or tuple
            size = augmenter_args.get("size")
            if size:
                if not isinstance(size, (list, tuple)) or len(size) != 2:
                    raise ValueError(
                        f"'size' must be a list or tuple of two numbers, but got {size}"
                    )
                min_scale, max_scale = size
                return {
                    "scale_range": (min_scale, max_scale),
                    "interpolation": 1,  # Linear interpolation
                    "p": 1.0,
                }
            else:
                return {"scale_range": (1.0, 1.0), "interpolation": 1, "p": 1.0}
        elif augmenter_type == "Affine":
            # Map rotation to a tuple and ensure p=1.0 to apply transformation
            rotate = augmenter_args.get("rotate", 0)
            if isinstance(rotate, list):
                rotate = tuple(rotate)
            elif isinstance(rotate, (int, float)):
                rotate = (float(rotate), float(rotate))
            augmenter_args["rotate"] = rotate

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. Give size as a two-number list or tuple, e.g. size: [1.0, 1.0] for no scaling
2. Understand the semantics: the values are scale factors, so use e.g. size: [0.8, 1.2] for 80-120% scaling
3. Omit size entirely to get the default no-scale mapping {scale_range: (1.0, 1.0)}
4. Validate the config programmatically before training: assert isinstance(size, (list, tuple)) and len(size) == 2

Example fix

# before
- { type: Resize, args: { size: 640 } }
# after
- { type: Resize, args: { size: [1.0, 1.0] } }

for entry in augmenter_args:
    if entry['type'] == 'Resize':
        size = entry.get('args', {}).get('size')
        if size is not None and (not isinstance(size, (list, tuple)) or len(size) != 2):
            raise SystemExit(f'Resize size must be [min_scale, max_scale], got {size}')

def is_valid_iaa_resize_size(size):
    import numbers
    return size is None or (
        isinstance(size, (list, tuple)) and len(size) == 2
        and all(isinstance(x, numbers.Number) for x in size)
    )

try:
    aug = IaaAugment(augmenter_args=augmenter_args)
except ValueError as e:
    raise ValueError(f'Invalid iaa Resize args: {e}') from e

When it happens

Trigger: Adding { type: Resize, args: { size: 640 } } or { type: Resize, args: { size: [640, 640, 3] } } to the augmenter_args list in a training config using IaaAugment.

Common situations: Writing imgaug-style configs where size is a single int (valid for some imgaug classes but not this mapper); copying an image shape [h, w, c] instead of a scale pair; note the two numbers are interpreted as min_scale/max_scale (fractions), so [640, 640] means a 640x scale, not pixels.

Related errors

• Unknown augmenter arg: {}
• OCR pipeline config text must decode to an object.
• OCR pipeline config must be an object or YAML text.
• ${modulePath}.model_dir must be null or an asset descriptor
• Unsupported pipeline_name "${pipelineName}". PaddleOCR.js cu

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/87682c74ced710f7. Report an issue: GitHub.