open-mmlab/mmdetection · error · RuntimeError

Invalid mode "{mode}". Only supports loss, predict and tenso

Error message

Invalid mode "{mode}". Only supports loss, predict and tensor mode

What it means

Raised by BaseMot.forward when the `mode` argument is anything other than 'loss', 'predict', or 'tensor'. MOT base models dispatch forward to loss/predict/_forward based on this mode string, mirroring mmdet's BaseDetector convention.

Source

Thrown at mmdet/models/mot/base.py:114

            data_samples (list[:obj:`TrackDataSample`], optional): The
                annotation data of every samples. Defaults to None.
            mode (str): Return what kind of value. Defaults to 'predict'.

        Returns:
            The return type depends on ``mode``.

            - If ``mode="tensor"``, return a tensor or a tuple of tensor.
            - If ``mode="predict"``, return a list of :obj:`TrackDataSample`.
            - If ``mode="loss"``, return a dict of tensor.
        """
        if mode == 'loss':
            return self.loss(inputs, data_samples, **kwargs)
        elif mode == 'predict':
            return self.predict(inputs, data_samples, **kwargs)
        elif mode == 'tensor':
            return self._forward(inputs, data_samples, **kwargs)
        else:
            raise RuntimeError(f'Invalid mode "{mode}". '
                               'Only supports loss, predict and tensor mode')

    @abstractmethod
    def loss(self, inputs: Dict[str, Tensor], data_samples: TrackSampleList,
             **kwargs) -> Union[dict, tuple]:
        """Calculate losses from a batch of inputs and data samples."""
        pass

    @abstractmethod
    def predict(self, inputs: Dict[str, Tensor], data_samples: TrackSampleList,
                **kwargs) -> TrackSampleList:
        """Predict results from a batch of inputs and data samples with post-
        processing."""
        pass

    def _forward(self,
                 inputs: Dict[str, Tensor],
                 data_samples: OptTrackSampleList = None,

View on GitHub (pinned to cfd5d3a985)

Solutions

1. Use mode='loss' when training, mode='predict' for inference returning DetDataSample, mode='tensor' for raw tensor outputs
2. If migrating old code, replace mode='train' with mode='loss' and mode='test' with mode='predict'

Example fix

# before
losses = model(imgs, batch_data_samples, mode='train')
# after
losses = model(imgs, batch_data_samples, mode='loss')

VALID = {'loss','predict','tensor'}
assert mode in VALID, f'mode must be one of {VALID}, got {mode!r}'

def is_valid_mode(m: str) -> bool: return m in {'loss','predict','tensor'}

When it happens

Trigger: Calling mot_model(imgs, data_samples, mode='val') or mode='inference', mode='test', or forgetting mode entirely in a custom loop where inputs happen to be a string.

Common situations: Custom test/training scripts that assume an older mmdet API (mode='train') or that pass a train/eval flag instead of the supported mode strings.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

• trackeval is not installed,please install it by: pip install
• metric must be a list or a str.
• metric {metric} is not supported.
• module must be a str or a list.
• _forward function (namely 'tensor' mode) is not supported no

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/56876190f8a47c6c. Report an issue: GitHub.