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 emitted by DetInferencer._load_weights_to_model when the loaded checkpoint's meta contains neither 'dataset_meta' nor legacy 'CLASSES', so COCO class names are assumed for labeling predictions.

Source

Thrown at mmdet/apis/det_inferencer.py:130

            cfg (Config or ConfigDict, optional): The loaded config.
        """

        if checkpoint is not None:
            _load_checkpoint_to_model(model, checkpoint)
            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.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')}
        else:
            warnings.warn('Checkpoint is not loaded, and the inference '
                          'result is calculated by the randomly initialized '
                          'model!')
            warnings.warn('weights is None, use COCO classes by default.')
            model.dataset_meta = {'classes': get_classes('coco')}

        # Priority:  args.palette -> config -> checkpoint
        if self.palette != 'none':
            model.dataset_meta['palette'] = self.palette
        else:
            test_dataset_cfg = copy.deepcopy(cfg.test_dataloader.dataset)
            # lazy init. We only need the metainfo.
            test_dataset_cfg['lazy_init'] = True
            metainfo = DATASETS.build(test_dataset_cfg).metainfo

View on GitHub (pinned to cfd5d3a985)

Solutions

1. Pass classes explicitly: DetInferencer(model, weights, classes=[...]) or set dataset_meta in the config
2. Use a checkpoint saved by mmdet >= 3.0 training (it embeds dataset_meta)
3. If the model really is COCO, ignore the warning

Example fix

// before
inferencer = DetInferencer(model=cfg, weights='converted.pth')
// after
inferencer = DetInferencer(model=cfg, weights='converted.pth', classes=['a','b','c'])

import torch
ckpt = torch.load('weights.pth', map_location='cpu')
meta = ckpt.get('meta', {})
if 'dataset_meta' not in meta and 'CLASSES' not in meta:
    print('checkpoint lacks class names; pass classes explicitly')

When it happens

Trigger: Running DetInferencer with a weights checkpoint whose meta dict lacks class info — typically checkpoints converted from other frameworks or from pre-3.x mmdet without meta fields.

Common situations: Using converted/foreign weights (e.g. torchvision or original-author released checkpoints) with DetInferencer; results still work but class labels may be wrong for non-COCO models.

Related errors

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

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/12214528b7569564. Report an issue: GitHub.