kornia.geometry.warp¶
-
class
HomographyWarper(height: int, width: int, mode: str = 'bilinear', padding_mode: str = 'zeros', normalized_coordinates: bool = True, align_corners: bool = False)[source]¶ Warp tensors by homographies.
\[X_{dst} = H_{src}^{\{dst\}} * X_{src}\]- Parameters
height (int) – The height of the destination tensor.
width (int) – The width of the destination tensor.
mode (str) – interpolation mode to calculate output values ‘bilinear’ | ‘nearest’. Default: ‘bilinear’.
padding_mode (str) – padding mode for outside grid values ‘zeros’ | ‘border’ | ‘reflection’. Default: ‘zeros’.
normalized_coordinates (bool) – wether to use a grid with normalized coordinates.
align_corners (bool) – interpolation flag. Default: False. See
https – //pytorch.org/docs/stable/nn.functional.html#torch.nn.functional.interpolate for detail
-
forward(patch_src: torch.Tensor, src_homo_dst: Optional[torch.Tensor] = None) → torch.Tensor[source]¶ Warp a tensor from source into reference frame.
- Parameters
patch_src (torch.Tensor) – The tensor to warp.
src_homo_dst (torch.Tensor, optional) – The homography or stack of homographies from destination to source. The homography assumes normalized coordinates [-1, 1] if normalized_coordinates is True. Default: None.
- Returns
Patch sampled at locations from source to destination.
- Return type
- Shape:
Input: \((N, C, H, W)\) and \((N, 3, 3)\)
Output: \((N, C, H, W)\)
Example
>>> input = torch.rand(1, 3, 32, 32) >>> homography = torch.eye(3).view(1, 3, 3) >>> warper = HomographyWarper(32, 32) >>> # without precomputing the warp >>> output = warper(input, homography) # NxCxHxW >>> # precomputing the warp >>> warper.precompute_warp_grid(homography) >>> output = warper(input) # NxCxHxW
-
precompute_warp_grid(src_homo_dst: torch.Tensor) → None[source]¶ Compute and store internaly the transformations of the points.
Useful when the same homography/homographies are reused.
- Parameters
src_homo_dst (torch.Tensor) – Homography or homographies (stacked) to transform all points in the grid. Shape of the homography has to be \((1, 3, 3)\) or \((N, 1, 3, 3)\). The homography assumes normalized coordinates [-1, 1] if normalized_coordinates is True.
-
class
DepthWarper(pinhole_dst: kornia.geometry.camera.pinhole.PinholeCamera, height: int, width: int, mode: str = 'bilinear', padding_mode: str = 'zeros', align_corners: bool = True)[source]¶ Warps a patch by depth.
\[ \begin{align}\begin{aligned}P_{src}^{\{dst\}} = K_{dst} * T_{src}^{\{dst\}}\\\begin{split}I_{src} = \\omega(I_{dst}, P_{src}^{\{dst\}}, D_{src})\end{split}\end{aligned}\end{align} \]- Parameters
pinholes_dst (PinholeCamera) – the pinhole models for the destination frame.
height (int) – the height of the image to warp.
width (int) – the width of the image to warp.
mode (str) – interpolation mode to calculate output values ‘bilinear’ | ‘nearest’. Default: ‘bilinear’.
padding_mode (str) – padding mode for outside grid values ‘zeros’ | ‘border’ | ‘reflection’. Default: ‘zeros’.
align_corners (bool) – interpolation flag. Default: True. See https://pytorch.org/docs/stable/nn.functional.html#torch.nn.functional.interpolate for detail
-
compute_projection_matrix(pinhole_src: kornia.geometry.camera.pinhole.PinholeCamera) → kornia.geometry.warp.depth_warper.DepthWarper[source]¶ Computes the projection matrix from the source to destination frame.
-
compute_subpixel_step() → torch.Tensor[source]¶ This computes the required inverse depth step to achieve sub pixel accurate sampling of the depth cost volume, per camera.
Szeliski, Richard, and Daniel Scharstein. “Symmetric sub-pixel stereo matching.” European Conference on Computer Vision. Springer Berlin Heidelberg, 2002.
-
forward(depth_src: torch.Tensor, patch_dst: torch.Tensor) → torch.Tensor[source]¶ Warps a tensor from destination frame to reference given the depth in the reference frame.
- Parameters
depth_src (torch.Tensor) – the depth in the reference frame. The tensor must have a shape \((B, 1, H, W)\).
patch_dst (torch.Tensor) – the patch in the destination frame. The tensor must have a shape \((B, C, H, W)\).
- Returns
the warped patch from destination frame to reference.
- Return type
- Shape:
Output: \((N, C, H, W)\) where C = number of channels.
Example
>>> # pinholes camera models >>> pinhole_dst = PinholeCamera(torch.randn(1, 4, 4), torch.randn(1, 4, 4), ... torch.tensor([32]), torch.tensor([32])) >>> pinhole_src = PinholeCamera(torch.randn(1, 4, 4), torch.randn(1, 4, 4), ... torch.tensor([32]), torch.tensor([32])) >>> # create the depth warper, compute the projection matrix >>> warper = DepthWarper(pinhole_dst, 32, 32) >>> _ = warper.compute_projection_matrix(pinhole_src) >>> # warp the destionation frame to reference by depth >>> depth_src = torch.ones(1, 1, 32, 32) # Nx1xHxW >>> image_dst = torch.rand(1, 3, 32, 32) # NxCxHxW >>> image_src = warper(depth_src, image_dst) # NxCxHxW
-
warp_grid(depth_src: torch.Tensor) → torch.Tensor[source]¶ Computes a grid for warping a given the depth from the reference pinhole camera.
The function compute_projection_matrix has to be called beforehand in order to have precomputed the relative projection matrices encoding the relative pose and the intrinsics between the reference and a non reference camera.
-
homography_warp(patch_src: torch.Tensor, src_homo_dst: torch.Tensor, dsize: Tuple[int, int], mode: str = 'bilinear', padding_mode: str = 'zeros', align_corners: bool = False, normalized_coordinates: bool = True) → torch.Tensor[source]¶ Warp image patchs or tensors by normalized 2D homographies.
See
HomographyWarperfor details.- Parameters
patch_src (torch.Tensor) – The image or tensor to warp. Should be from source of shape \((N, C, H, W)\).
src_homo_dst (torch.Tensor) – The homography or stack of homographies from destination to source of shape \((N, 3, 3)\).
dsize (Tuple[int, int]) – The height and width of the image to warp.
mode (str) – interpolation mode to calculate output values ‘bilinear’ | ‘nearest’. Default: ‘bilinear’.
padding_mode (str) – padding mode for outside grid values ‘zeros’ | ‘border’ | ‘reflection’. Default: ‘zeros’.
align_corners (bool) – interpolation flag. Default: False. See https://pytorch.org/docs/stable/nn.functional.html#torch.nn.functional.interpolate for detail
normalized_coordinates (bool) – Whether the homography assumes [-1, 1] normalized coordinates or not.
- Returns
Patch sampled at locations from source to destination.
- Return type
Example
>>> input = torch.rand(1, 3, 32, 32) >>> homography = torch.eye(3).view(1, 3, 3) >>> output = homography_warp(input, homography, (32, 32))
-
depth_warp(pinhole_dst: kornia.geometry.camera.pinhole.PinholeCamera, pinhole_src: kornia.geometry.camera.pinhole.PinholeCamera, depth_src: torch.Tensor, patch_dst: torch.Tensor, height: int, width: int, align_corners: bool = True)[source]¶ Function that warps a tensor from destination frame to reference given the depth in the reference frame.
See
DepthWarperfor details.Example
>>> # pinholes camera models >>> pinhole_dst = PinholeCamera(torch.randn(1, 4, 4), torch.randn(1, 4, 4), ... torch.tensor([32]), torch.tensor([32])) >>> pinhole_src = PinholeCamera(torch.randn(1, 4, 4), torch.randn(1, 4, 4), ... torch.tensor([32]), torch.tensor([32])) >>> # warp the destionation frame to reference by depth >>> depth_src = torch.ones(1, 1, 32, 32) # Nx1xHxW >>> image_dst = torch.rand(1, 3, 32, 32) # NxCxHxW >>> image_src = depth_warp(pinhole_dst, pinhole_src, depth_src, image_dst, 32, 32) # NxCxHxW
-
warp_grid(grid: torch.Tensor, src_homo_dst: torch.Tensor) → torch.Tensor[source]¶ Compute the grid to warp the coordinates grid by the homography/ies.
- Parameters
grid – Unwrapped grid of the shape \((1, N, W, 2)\).
src_homo_dst (torch.Tensor) – Homography or homographies (stacked) to transform all points in the grid. Shape of the homography has to be \((1, 3, 3)\) or \((N, 1, 3, 3)\).
- Returns
the transformed grid of shape \((N, H, W, 2)\).
- Return type
-
normalize_homography(dst_pix_trans_src_pix: torch.Tensor, dsize_src: Tuple[int, int], dsize_dst: Tuple[int, int]) → torch.Tensor[source]¶ Normalize a given homography in pixels to [-1, 1].
- Parameters
dst_pix_trans_src_pix (torch.Tensor) – homography/ies from source to destination to be normalized. \((B, 3, 3)\)
dsize_src (tuple) – size of the source image (height, width).
dsize_dst (tuple) – size of the destination image (height, width).
- Returns
the normalized homography of shape \((B, 3, 3)\).
- Return type
-
normal_transform_pixel(height: int, width: int, eps: float = 1e-14, device: Optional[torch.device] = None, dtype: Optional[torch.dtype] = None) → torch.Tensor[source]¶ Compute the normalization matrix from image size in pixels to [-1, 1].
- Parameters
- Returns
normalized transform with shape \((1, 3, 3)\).
- Return type