``single_level_grid_anchors`` would be deprecated soon. Plea

Error message

``single_level_grid_anchors`` would be deprecated soon. Please use ``single_level_grid_priors``

What it means

Deprecation warning from AnchorGenerator.single_level_grid_anchors: the per-level anchors API is superseded by single_level_grid_priors. It is still called internally by grid_anchors but direct calls will break once removed.

Source

Thrown at mmdet/models/task_modules/prior_generators/anchor_generator.py:392

                                  device: DeviceType = 'cuda') -> Tensor:
        """Generate grid anchors of a single level.

        Note:
            This function is usually called by method ``self.grid_anchors``.

        Args:
            base_anchors (torch.Tensor): The base anchors of a feature grid.
            featmap_size (tuple[int]): Size of the feature maps.
            stride (tuple[int, int]): Stride of the feature map in order
                (w, h). Defaults to (16, 16).
            device (str | torch.device): Device the tensor will be put on.
                Defaults to 'cuda'.

        Returns:
            torch.Tensor: Anchors in the overall feature maps.
        """

        warnings.warn(
            '``single_level_grid_anchors`` would be deprecated soon. '
            'Please use ``single_level_grid_priors`` ')

        # keep featmap_size as Tensor instead of int, so that we
        # can convert to ONNX correctly
        feat_h, feat_w = featmap_size
        shift_x = torch.arange(0, feat_w, device=device) * stride[0]
        shift_y = torch.arange(0, feat_h, device=device) * stride[1]

        shift_xx, shift_yy = self._meshgrid(shift_x, shift_y)
        shifts = torch.stack([shift_xx, shift_yy, shift_xx, shift_yy], dim=-1)
        shifts = shifts.type_as(base_anchors)
        # first feat_w elements correspond to the first row of shifts
        # add A anchors (1, A, 4) to K shifts (K, 1, 4) to get
        # shifted anchors (K, A, 4), reshape to (K*A, 4)

        all_anchors = base_anchors[None, :, :] + shifts[:, None, :]
        all_anchors = all_anchors.view(-1, 4)

View on GitHub (pinned to cfd5d3a985)

Solutions

Call single_level_grid_priors(...) instead
Prefer grid_priors for whole-pyramid generation rather than per-level calls
Update transitive callers to the priors API to silence both warnings

Example fix

# before
anchors = generator.single_level_grid_anchors(featmap_size, stride, device=device)
# after
priors = generator.single_level_grid_priors(featmap_size, stride, device=device)

Defensive patterns

Strategy: type-guard

Validate before calling

assert hasattr(gen, 'single_level_grid_priors'), 'use single_level_grid_priors'

Type guard

def has_priors_api(gen) -> bool:
    return hasattr(gen, 'single_level_grid_priors')

Prevention

Call single_level_grid_priors in custom heads
Avoid per-level anchor code; prefer grid_priors

When it happens

Trigger: Directly invoking single_level_grid_anchors(featmap_size, stride, device) in a custom head or test; also triggered transitively through grid_anchors.

Common situations: Custom anchor-based heads computing anchors level by level; code migration during the anchors→priors rename.

Related errors

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/bfd17d669db548c5. Report an issue: GitHub.