roboflow/supervision · error · ValueError

`slice_wh` must be a positive integer. Received: {slice_wh}

Error message

`slice_wh` must be a positive integer. Received: {slice_wh}

What it means

Raised by InferenceSlicer._normalize_slice_wh when slice_wh is an int that is zero or negative. The int form means 'square slices of this pixel size'; a non-positive size cannot tile any image, so construction fails.

Source

Thrown at src/supervision/detection/tools/inference_slicer.py:648

                                "full-resolution image.",
                                category=SupervisionWarnings,
                                stacklevel=2,
                            )

        return [
            move_detections(
                detections=det, offset=offset[:2], resolution_wh=resolution_wh
            )
            for det, offset in zip(detections_in_slices, offsets)
        ]

    @staticmethod
    def _normalize_slice_wh(
        slice_wh: int | tuple[int, int],
    ) -> tuple[int, int]:
        if isinstance(slice_wh, int):
            if slice_wh <= 0:
                raise ValueError(
                    f"`slice_wh` must be a positive integer. Received: {slice_wh}"
                )
            return slice_wh, slice_wh

        if isinstance(slice_wh, tuple) and len(slice_wh) == 2:
            width, height = slice_wh
            if width <= 0 or height <= 0:
                raise ValueError(
                    f"`slice_wh` values must be positive. Received: {slice_wh}"
                )
            return width, height

        raise ValueError(
            "`slice_wh` must be an int or a tuple of two positive integers "
            "(slice_w, slice_h). "
            f"Received: {slice_wh}"
        )

View on GitHub (pinned to 7f254d9784)

Solutions

1. Pass a positive pixel size, e.g. slice_wh=512.
2. Guard computed sizes: slice_wh = max(1, computed) — though realistically pick a real tile size like 320-1024.
3. Validate required config fields at load time instead of relying on constructor failure.

Example fix

# before
slicer = sv.InferenceSlicer(callback=cb, slice_wh=cfg.get('slice', 0))

# after
slice_size = cfg.get('slice') or 512
slicer = sv.InferenceSlicer(callback=cb, slice_size if isinstance(slice_size, tuple) else int(slice_size))

slice_wh = int(cfg.get('slice_wh') or 512)
assert slice_wh > 0, 'slice_wh must be a positive pixel size'
slicer = sv.InferenceSlicer(callback=cb, slice_wh=slice_wh)

def is_positive_int(v) -> bool:
    return isinstance(v, int) and v > 0

When it happens

Trigger: Passing slice_wh=0 or a negative int, typically from a computed value (e.g. target_size // scale that floors to 0) or an unset config default of 0.

Common situations: Downscaling math that produces 0 for very large divisors; config schemas where 0 is the 'unset' sentinel; CLI defaults leaking through.

Related errors

• `slice_wh` must be an int or a tuple of two positive integer
• `overlap_wh` must be an int or a tuple of two non negative i
• `thread_workers` must be a positive integer. Received: {thre
• `batch_size` must be a positive integer. Received: {batch_si
• `slice_wh` values must be positive. Received: {slice_wh}

AI-assisted analysis of roboflow/supervision@7f254d9784 (2026-08-15). Data as JSON: /api/errors/ebb819808ada8cbc. Report an issue: GitHub.