Match a pair of images (image0, image1) with an extractor and matcher

Multi-head scaled dot-product attention Args: q: [N, L, D] k: [N, S, D] v: [N, S, D] Returns: out: (N, L, D)

Normalize the image tensor and reorder the dimensions.

Computes the plucker coordinates from batched cam2world & intrinsics matrices, as well as pixel coordinates c2w: (B, 4, 4) intrinsics:

Converts 6D rotation representation by Zhou et al. [1] to rotation matrix using Gram--Schmidt orthogonalisation per Section B of [1].

Remove batch dimension from elements in data

Read an image from path as RGB or grayscale

Resize an image to a fixed size, or according to max or min edge.

Restore the parameters stored with the `store` method. Useful to validate the model with EMA parameters without affecting the

Detect keypoints using OpenCV Detector. Optionally, perform description. Args: features: OpenCV based keypoints detector and desc

Interpolate descriptors at keypoint locations

Save an image. Assumed to be in range 0-1.

create the log assignment matrix from logits and similarity

Fast Non-maximum suppression to remove nearby points

Fast Non-maximum suppression to remove nearby points

r"""SFNet: Learning Object-aware Semantic Flow (Lee et al.)

Save the current parameters for restoring later. Args: parameters: Iterable of `torch.nn.Parameter`; the parameters to b

Transform points from 3D camera coordinates to 3D world coordinates.

Transform points from 3D world coordinates to 3D camera coordinates.

feature1: [B, C, H, W] intrinsics: [B, 3, 3] pose: [B, 4, 4] depth: [B, D, H, W]

Zero out the parameters of a module and return it.

Resize and preprocess an image, return image and resize scale

The harmonic embedding layer supports the classical Nerf positional encoding described in `NeRF <https://arxiv.org/abs/2003.0