DeprecationWarning: bboxes is deprecated, please use "priors

Error message

DeprecationWarning: bboxes is deprecated, please use "priors" instead

What it means

Deprecation warning from SamplingResult.bboxes property: the bboxes/pos_bboxes/neg_bboxes attributes were renamed to priors/pos_priors/neg_priors because sampled 'boxes' are actually the prior anchors. The property just forwards to self.priors and will be removed.

Source

Thrown at mmdet/models/task_modules/samplers/sampling_result.py:126

        box_dim = gt_bboxes.box_dim if isinstance(gt_bboxes, BaseBoxes) else 4
        if gt_bboxes.numel() == 0:
            # hack for index error case
            assert self.pos_assigned_gt_inds.numel() == 0
            self.pos_gt_bboxes = gt_bboxes.view(-1, box_dim)
        else:
            if len(gt_bboxes.shape) < 2:
                gt_bboxes = gt_bboxes.view(-1, box_dim)
            self.pos_gt_bboxes = gt_bboxes[self.pos_assigned_gt_inds.long()]

    @property
    def priors(self):
        """torch.Tensor: concatenated positive and negative priors"""
        return cat_boxes([self.pos_priors, self.neg_priors])

    @property
    def bboxes(self):
        """torch.Tensor: concatenated positive and negative boxes"""
        warnings.warn('DeprecationWarning: bboxes is deprecated, '
                      'please use "priors" instead')
        return self.priors

    @property
    def pos_bboxes(self):
        warnings.warn('DeprecationWarning: pos_bboxes is deprecated, '
                      'please use "pos_priors" instead')
        return self.pos_priors

    @property
    def neg_bboxes(self):
        warnings.warn('DeprecationWarning: neg_bboxes is deprecated, '
                      'please use "neg_priors" instead')
        return self.neg_priors

    def to(self, device):
        """Change the device of the data inplace.

View on GitHub (pinned to cfd5d3a985)

Solutions

Use sampling_result.priors instead of .bboxes
Similarly switch pos_bboxes→pos_priors, neg_bboxes→neg_priors (warnings 317/318)
Grep codebase for '_bboxes' usages on SamplingResult objects

Example fix

# before
rois = sampling_result.bboxes
# after
rois = sampling_result.priors

Defensive patterns

Strategy: type-guard

Validate before calling

assert hasattr(result, 'priors'), 'SamplingResult lacks priors; upgrade mmdet'

Type guard

def get_priors(result):
    return result.priors if hasattr(result, 'priors') else result.bboxes

Prevention

Consume SamplingResult via priors/pos_priors/neg_priors
Feature-detect attributes when supporting multiple mmdet versions

When it happens

Trigger: Accessing sampling_result.bboxes in custom sampler/roi-head code or tests; anything consuming SamplingResult output written against the old attribute names.

Common situations: Custom RoI heads or debugging code reading result.bboxes / result.pos_bboxes after mmdet renamed sampling fields to priors.

Related errors

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/9c8e1610a5b669c7. Report an issue: GitHub.