open-mmlab/mmdetection · warning

dataset_meta or class names are not saved in the checkpoint'

Error message

dataset_meta or class names are not saved in the checkpoint's meta data, use COCO classes by default.

What it means

Warning from init_detector: the checkpoint loaded successfully but its meta dict has neither 'dataset_meta' nor legacy 'CLASSES', so COCO classes are the default labels.

Source

Thrown at mmdet/apis/inference.py:90

    else:
        checkpoint = load_checkpoint(model, checkpoint, map_location='cpu')
        # Weights converted from elsewhere may not have meta fields.
        checkpoint_meta = checkpoint.get('meta', {})

        # save the dataset_meta in the model for convenience
        if 'dataset_meta' in checkpoint_meta:
            # mmdet 3.x, all keys should be lowercase
            model.dataset_meta = {
                k.lower(): v
                for k, v in checkpoint_meta['dataset_meta'].items()
            }
        elif 'CLASSES' in checkpoint_meta:
            # < mmdet 3.x
            classes = checkpoint_meta['CLASSES']
            model.dataset_meta = {'classes': classes}
        else:
            warnings.simplefilter('once')
            warnings.warn(
                'dataset_meta or class names are not saved in the '
                'checkpoint\'s meta data, use COCO classes by default.')
            model.dataset_meta = {'classes': get_classes('coco')}

    # Priority:  args.palette -> config -> checkpoint
    if palette != 'none':
        model.dataset_meta['palette'] = palette
    else:
        test_dataset_cfg = copy.deepcopy(config.test_dataloader.dataset)
        # lazy init. We only need the metainfo.
        test_dataset_cfg['lazy_init'] = True
        metainfo = DATASETS.build(test_dataset_cfg).metainfo
        cfg_palette = metainfo.get('palette', None)
        if cfg_palette is not None:
            model.dataset_meta['palette'] = cfg_palette
        else:
            if 'palette' not in model.dataset_meta:
                warnings.warn(

View on GitHub (pinned to cfd5d3a985)

Solutions

1. Prefer checkpoints from official mmdet training
2. Set classes via config: model.test_cfg / dataset metainfo, or manually assign model.dataset_meta after init
3. Patch the checkpoint meta: ckpt['meta']['dataset_meta'] = {'classes': [...]} then re-save

Example fix

// before
model = init_detector(cfg, 'converted.pth')
// after
model = init_detector(cfg, 'converted.pth')
model.dataset_meta = {'classes': my_class_list}

import torch
meta = torch.load(ckpt, map_location='cpu').get('meta', {})
assert 'dataset_meta' in meta or 'CLASSES' in meta or explicit_classes, 'labels will default to COCO'

When it happens

Trigger: init_detector(cfg, 'weights.pth') where weights.pth was converted from another framework or stripped of meta.

Common situations: Using mmengine-converted or third-party weights; predictions are fine but class names/ids may mismatch for non-COCO-trained models.

Related errors

• dataset_meta or class names are not saved in the checkpoint'
• checkpoint is None, use COCO classes by default.
• Checkpoint is not loaded, and the inference result is calcul
• weights is None, use COCO classes by default.
• palette does not exist, random is used by default. You can a

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/e99cf761adeafeb2. Report an issue: GitHub.