facebookresearch/detectron2 · error · ValueError

One annotation of image {image_id} contains empty 'bbox' val

Error message

One annotation of image {image_id} contains empty 'bbox' value! This json does not have valid COCO format.

What it means

load_coco_json iterates annotations and rejects any whose 'bbox' list is empty (len 0), because an empty box is meaningless for training and indicates a malformed COCO json.

Source

Thrown at detectron2/data/datasets/coco.py:180

        record["width"] = img_dict["width"]
        image_id = record["image_id"] = img_dict["id"]

        objs = []
        for anno in anno_dict_list:
            # Check that the image_id in this annotation is the same as
            # the image_id we're looking at.
            # This fails only when the data parsing logic or the annotation file is buggy.

            # The original COCO valminusminival2014 & minival2014 annotation files
            # actually contains bugs that, together with certain ways of using COCO API,
            # can trigger this assertion.
            assert anno["image_id"] == image_id

            assert anno.get("ignore", 0) == 0, '"ignore" in COCO json file is not supported.'

            obj = {key: anno[key] for key in ann_keys if key in anno}
            if "bbox" in obj and len(obj["bbox"]) == 0:
                raise ValueError(
                    f"One annotation of image {image_id} contains empty 'bbox' value! "
                    "This json does not have valid COCO format."
                )

            segm = anno.get("segmentation", None)
            if segm:  # either list[list[float]] or dict(RLE)
                if isinstance(segm, dict):
                    if isinstance(segm["counts"], list):
                        # convert to compressed RLE
                        segm = mask_util.frPyObjects(segm, *segm["size"])
                else:
                    # filter out invalid polygons (< 3 points)
                    segm = [poly for poly in segm if len(poly) % 2 == 0 and len(poly) >= 6]
                    if len(segm) == 0:
                        num_instances_without_valid_segmentation += 1
                        continue  # ignore this instance
                obj["segmentation"] = segm

View on GitHub (pinned to a2f4a8771a)

Solutions

1. Clean the json: drop annotations with empty or invalid bbox before registering
2. Fix the converter to skip objects without a valid 4-element bbox
3. Or patch upstream to tolerate and skip, though cleaning the data is preferred

Example fix

# before
for ann in coco_json["annotations"]:
    pass  # empty bboxes present
# after
import json
d = json.load(open('instances.json'))
d["annotations"] = [a for a in d["annotations"] if a.get("bbox") and len(a["bbox"]) == 4]
json.dump(d, open('instances_clean.json', 'w'))

import json
with open('instances.json') as f:
    d = json.load(f)
bad = [a for a in d['annotations'] if not a.get('bbox')]
assert not bad, f'{len(bad)} annotations have empty bbox'

def has_valid_bbox(ann) -> bool:
    b = ann.get('bbox')
    return isinstance(b, (list, tuple)) and len(b) == 4 and all(isinstance(v, (int, float)) for v in b)

try:
    d = load_coco_json(json_file, img_root, 'mydata')
except ValueError as e:
    raise SystemExit(f'malformed COCO json: {e}') from e

When it happens

Trigger: Loading a COCO-format json (register_coco_instances / load_coco_json) where some annotation has "bbox": [] — often produced by converters that emit empty bboxes for degenerate objects instead of omitting the annotation.

Common situations: Third-party conversion scripts (VOC/CVAT/labelme -> COCO) writing empty bboxes; datasets with fully-cropped-out or zero-size objects; hand-edited jsons.

Related errors

• Encountered category_id={annotation_category_id} but this id
• Cannot match one checkpoint key to multiple keys in the mode
• Class with @configurable must have a 'from_config' classmeth
• {name} must take 'cfg' as the first argument!
• target of LazyCall must be a callable or defines a callable!

AI-assisted analysis of facebookresearch/detectron2@a2f4a8771a (2026-08-27). Data as JSON: /api/errors/e3c0a4378f6ff42d. Report an issue: GitHub.