PaddlePaddle/PaddleOCR · error · Exception

Unsupported interpolation type !!!

Error message

Unsupported interpolation type !!!

What it means

This grayscale-recognition image-normalization class accepts only OpenCV interpolation codes 0 (NEAREST), 1 (LINEAR), 2 (CUBIC), and 3 (AREA) passed as plain ints. Any other value raises 'Unsupported interpolation type !!!' at pipeline construction time.

Source

Thrown at ppocr/data/imaug/rec_img_aug.py:364

        return data


class RFLRecResizeImg(object):
    def __init__(self, image_shape, padding=True, interpolation=1, **kwargs):
        self.image_shape = image_shape
        self.padding = padding

        self.interpolation = interpolation
        if self.interpolation == 0:
            self.interpolation = cv2.INTER_NEAREST
        elif self.interpolation == 1:
            self.interpolation = cv2.INTER_LINEAR
        elif self.interpolation == 2:
            self.interpolation = cv2.INTER_CUBIC
        elif self.interpolation == 3:
            self.interpolation = cv2.INTER_AREA
        else:
            raise Exception("Unsupported interpolation type !!!")

    def __call__(self, data):
        img = data["image"]
        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        norm_img, valid_ratio = resize_norm_img(
            img, self.image_shape, self.padding, self.interpolation
        )
        data["image"] = norm_img
        data["valid_ratio"] = valid_ratio
        if "iluvatar_gpu" in get_device():
            data["valid_ratio"] = np.float32(valid_ratio)
        return data


class SRNRecResizeImg(object):
    def __init__(self, image_shape, num_heads, max_text_length, **kwargs):
        self.image_shape = image_shape
        self.num_heads = num_heads

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. Set interpolation to one of 0/1/2/3 in the yml (e.g. interpolation: 1 for LINEAR)
2. If you have a cv2 constant in Python, pass int(cv2.INTER_LINEAR) not the flag name string
3. For Lanczos-style quality use 2 (CUBIC) as the closest supported option
4. Add a startup assertion in custom code: assert op.interpolation in (0, 1, 2, 3)

Example fix

# before
interpolation: 4
# after
interpolation: 2

iv = op_cfg.get('interpolation', 1)
if iv not in (0, 1, 2, 3):
    raise SystemExit(f'interpolation must be 0|1|2|3, got {iv!r}')

def is_valid_interp(v) -> bool:
    return isinstance(v, int) and not isinstance(v, bool) and 0 <= v <= 3

try:
    op = RecResizeNorm(**op_cfg)
except Exception as e:
    if 'Unsupported interpolation' in str(e):
        raise ValueError(f"Use interpolation 0-3 (NEAREST/LINEAR/CUBIC/AREA), got {op_cfg.get('interpolation')}") from e
    raise

When it happens

Trigger: Setting interpolation: 4 (or any int outside 0-3), a string like 'linear', or a cv2 enum constant (cv2.INTER_LANCZOS4 = 4) in the rec preprocessing config for this op.

Common situations: Using Lanczos4 (4) or other OpenCV flags unsupported by this mapper; passing a cv2 constant object instead of its int value; configs ported from torch/torchvision where strings like 'bilinear' are the convention.

Related errors

• RecResizeImg.image_shape is required in rec inference.yml
• Unexpected recognition channels: ${String(channels)}
• Type of target_size is invalid. Now is {}
• not support limit type, image
• {} is not supported in MultiLoss yet

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/e59a4e917ef30e08. Report an issue: GitHub.