Comfy-Org/ComfyUI · error · ValueError

Invalid rotation matrix shape {matrix.shape}.

Error message

Invalid rotation matrix shape {matrix.shape}.

What it means

mat_to_quat() converts rotation matrices to xyzw quaternions and requires the input's last two dimensions to be exactly 3x3. Any other trailing shape raises ValueError with the received shape. It is a hard precondition of the conversion math, which unbinds exactly nine matrix entries.

Source

Thrown at comfy/ldm/depth_anything_3/transform.py:80

            1 - two_s * (j * j + k * k),
            two_s * (i * j - k * r),
            two_s * (i * k + j * r),
            two_s * (i * j + k * r),
            1 - two_s * (i * i + k * k),
            two_s * (j * k - i * r),
            two_s * (i * k - j * r),
            two_s * (j * k + i * r),
            1 - two_s * (i * i + j * j),
        ),
        -1,
    )
    return o.reshape(quaternions.shape[:-1] + (3, 3))


def mat_to_quat(matrix: torch.Tensor) -> torch.Tensor:
    """Convert (...,3,3) rotation matrices to quaternions (xyzw)."""
    if matrix.size(-1) != 3 or matrix.size(-2) != 3:
        raise ValueError(f"Invalid rotation matrix shape {matrix.shape}.")

    batch_dim = matrix.shape[:-2]
    m00, m01, m02, m10, m11, m12, m20, m21, m22 = torch.unbind(
        matrix.reshape(batch_dim + (9,)), dim=-1
    )

    q_abs = _sqrt_positive_part(
        torch.stack(
            [
                1.0 + m00 + m11 + m22,
                1.0 + m00 - m11 - m22,
                1.0 - m00 + m11 - m22,
                1.0 - m00 - m11 + m22,
            ],
            dim=-1,
        )
    )

View on GitHub (pinned to 1c6d8d45b3)

Solutions

1. Slice the rotation block first: mat_to_quat(pose[..., :3, :3]).
2. Verify with matrix.shape[-2:] == (3, 3) before the call when the source of the tensor is uncertain.
3. If input is (3,4) extrinsics, drop the translation column before converting.

Example fix

# before
q = mat_to_quat(pose_4x4)

# after
q = mat_to_quat(pose_4x4[..., :3, :3])

assert matrix.shape[-2:] == (3, 3), f"need (...,3,3), got {tuple(matrix.shape)}"

def is_rotation_matrix_tensor(m: "torch.Tensor") -> bool:
    return m.ndim >= 2 and m.shape[-2:] == (3, 3)

When it happens

Trigger: Calling mat_to_quat on 4x4 pose matrices (forgetting to slice [...,:3,:3]), on (...,3) Euler angles, on batched tensors whose reshape earlier collapsed the matrix dims, or on transposed translation-augmented (3,4) tensors.

Common situations: Camera-pose pipelines that store homogeneous 4x4 matrices and pass them straight through; mixing up quaternion-to-matrix (output 3x3) and matrix-to-quaternion (input must be 3x3) call directions.

Related errors

• Invalid affine shape: {ext.shape}
• Unexpected token width: {out_x.shape[-1]}
• Unknown DINOv2 backbone variant: {backbone_name!r}
• Unsupported process_res_method: {method}
• At least 4 points are required to compute a homography.

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/e0ed41a7d2424b20. Report an issue: GitHub.