Function warp_perspective

imperative/python/megengine/functional/vision.py:395–464

r"""Applies perspective transformation to batched 2D images. A perspective transformation is a projection of a image onto a new view plane. The input images are transformed to the output images by the transformation matrix: .. math:: \text{output}(n, c, h, w) = \text{input} \le

(
    inp: Tensor,
    mat: Tensor,
    out_shape: Union[Tuple[int, int], int, Tensor],
    mat_idx: Optional[Union[Iterable[int], Tensor]] = None,
    border_mode: str = "replicate",
    border_val: float = 0.0,
    format: str = "NCHW",
    interp_mode: str = "linear",
)

def warp_perspective(
    inp: Tensor,
    mat: Tensor,
    out_shape: Union[Tuple[int, int], int, Tensor],
    mat_idx: Optional[Union[Iterable[int], Tensor]] = None,
    border_mode: str = "replicate",
    border_val: float = 0.0,
    format: str = "NCHW",
    interp_mode: str = "linear",
) -> Tensor:
    r"""Applies perspective transformation to batched 2D images. A perspective transformation is a projection of a image onto a new view plane.

    The input images are transformed to the output images by the transformation matrix:

    .. math::
            \text{output}(n, c, h, w) = \text{input} \left( n, c,
                \frac{M_{00}w + M_{01}h + M_{02}}{M_{20}w + M_{21}h + M_{22}},
                \frac{M_{10}w + M_{11}h + M_{12}}{M_{20}w + M_{21}h + M_{22}}
                \right)

    Optionally, we can set ``mat_idx`` to assign different transformations to the same image,
    otherwise the input images and transformations should be one-to-one correnspondence.

    Args:
        inp: input image.
        mat: ``(batch, 3, 3)`` transformation matrix.
        out_shape: ``(h, w)`` size of the output image.
        mat_idx: image batch idx assigned to each matrix. Default: None
        border_mode: pixel extrapolation method.
            Default: "replicate". Currently also support "constant", "reflect",
            "reflect_101", "wrap".
        border_val: value used in case of a constant border. Default: 0
        format: NHWC" is also supported. Default: "NCHW".
        interp_mode: interpolation methods.
            Default: "linear". Currently only support "linear" mode.

    Returns:
        output tensor.

    Note:
        The transformation matrix is the inverse of that used by ``cv2.warpPerspective``.

    Examples:
        >>> import numpy as np
        >>> inp_shape = (1, 1, 4, 4)
        >>> x = Tensor(np.arange(16, dtype=np.float32).reshape(inp_shape))
        >>> M_shape = (1, 3, 3)
        >>> # M defines a translation: dst(1, 1, h, w) = rst(1, 1, h+1, w+1)
        >>> M = Tensor(np.array([[1., 0., 1.],
        ...                      [0., 1., 1.],
        ...                      [0., 0., 1.]], dtype=np.float32).reshape(M_shape))
        >>> out = F.vision.warp_perspective(x, M, (2, 2))
        >>> out.numpy()
        array([[[[ 5.,  6.],
                 [ 9., 10.]]]], dtype=float32)
    """
    if inp.dtype == np.float32:
        mat = mat.astype("float32")