WZMIAOMIAO/deep-learning-for-image-processing · error · ValueError

All bounding boxes should have positive height and width. Fo

Error message

All bounding boxes should have positive height and width. Found invalid box {} for target at index {}.

What it means

During training forward, RetinaNet checks every target for degenerate boxes where x2 <= x1 or y2 <= y1 (non-positive width/height) and raises ValueError identifying the first invalid box and the target index. Degenerate boxes break IoU computation and anchor assignment downstream.

Source

Thrown at pytorch_object_detection/retinaNet/network_files/retinanet.py:490

        for img in images:
            val = img.shape[-2:]
            assert len(val) == 2
            original_img_sizes.append((val[0], val[1]))  # h, w

        # transform the input
        images, targets = self.transform(images, targets)

        # Check for degenerate boxes
        # TODO: Move this to a function
        if targets is not None:
            for target_idx, target in enumerate(targets):
                boxes = target["boxes"]
                degenerate_boxes = boxes[:, 2:] <= boxes[:, :2]
                if degenerate_boxes.any():
                    # print the first degenerate box
                    bb_idx = torch.where(degenerate_boxes.any(dim=1))[0][0]
                    degen_bb: List[float] = boxes[bb_idx].tolist()
                    raise ValueError("All bounding boxes should have positive height and width."
                                     " Found invalid box {} for target at index {}."
                                     .format(degen_bb, target_idx))

        # get the features from the backbone
        features = self.backbone(images.tensors)
        if isinstance(features, torch.Tensor):
            features = OrderedDict([("0", features)])

        features = list(features.values())

        # compute the retinanet heads outputs using the features
        head_outputs = self.head(features)

        # create the set of anchors
        anchors = self.anchor_generator(images, features)

        losses = {}
        detections: List[Dict[str, Tensor]] = []

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Sanitize annotations: drop or fix boxes where x2 <= x1 or y2 <= y1 before training.
2. Verify coordinate format is xyxy (top-left, bottom-right), not xywh or cxcywh.
3. Clip boxes then re-filter degenerates after clipping to image boundaries.
4. Assert in the Dataset: assert (boxes[:, 2:] > boxes[:, :2]).all() before returning the target.

Example fix

// before
boxes = torch.as_tensor(raw_boxes, dtype=torch.float32)
target = {"boxes": boxes, "labels": labels}
// after
boxes = torch.as_tensor(raw_boxes, dtype=torch.float32)
keep = (boxes[:, 2:] > boxes[:, :2]).all(dim=1)
boxes, labels = boxes[keep], labels[keep]  # drop degenerate boxes
target = {"boxes": boxes, "labels": labels}

import torch
for i, t in enumerate(targets):
    boxes = t["boxes"]
    if ((boxes[:, 2:] <= boxes[:, :2]).any()):
        bad = boxes[(boxes[:, 2:] <= boxes[:, :2]).any(dim=1)]
        print(f"target {i} has degenerate boxes: {bad.tolist()}")

def boxes_are_valid(boxes) -> bool:
    import torch
    if not isinstance(boxes, torch.Tensor) or boxes.numel() == 0:
        return True
    return bool((boxes[:, 2:] > boxes[:, :2]).all())

try:
    outputs = model(images, targets)
except ValueError as e:
    if "positive height and width" in str(e):
        targets = [sanitize_boxes(t) for t in targets]  # drop degenerate rows
        outputs = model(images, targets)
    else:
        raise

When it happens

Trigger: Annotations containing zero-area boxes (x1==x2 or y1==y2), negative width/height from wrong coordinate order (x_min > x_max), or boxes in [cx, cy, w, h] format passed as xyxy; coordinates flipped after resize/clipping transforms.

Common situations: Datasets converted between xyxy/xywh formats incorrectly; annotation tools exporting empty boxes for invisible objects; clip-to-image operations that collapse boxes to zero width at image borders; single-point objects (keypoints) labeled as boxes.

Related errors

• No ground-truth boxes available for one of the images during
• No proposal boxes available for one of the images during tra
• backbone should contain an attribute out_channels specifying
• In training mode, targets should be passed
• Expected target boxes to be a tensorof shape [N, 4], got {:}

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/80f57c8bf1e7b402. Report an issue: GitHub.