Initialize parts of decoder that are dependent on dimension of encoder tokens. Should be called when setting up MultiMAE. :p

Check if a viewpoint is valid and overlaps significantly with a reference one.

Enhanced load_model function with HuggingFace support

Loads a .npz file containing: - images_jpeg_bytes: compressed frames - tracks_XYZ: shape (T, N, 3), camera-space 3D points - fx

renorm pointmaps pts1_seq with pts2_seq's scale with norm_mode

Args: x3d (torch.Tensor): Shape (*, n, 3) pose (torch.Tensor | None): Shape (*, 4) detach_transformation (bool) Retu

Project points from world coordinates to 2D image plane using camera extrinsics and intrinsics. Args: - points_3d (T, N, 3 np.array)

Sample a random viewpoint using the navmesh

This function disables printing when not in master process

Args: x (torch.Tensor): shape (*, 3) Returns: torch.Tensor: (*, 3, 3), skew symmetric matrices

Calculate trajectory loss, including optimizing a scale factor by aligning predicted trajectories with ground truth trajectories. Args:

Convert Euclidean distances (sqrt(x^2 + y^2 + z^2)) in depthmap to actual Z values in camera coordinate system. Parameters: - depthm

Output a (H,W,2) array of int32 with output[j,i,0] = i + origin[0] output[j,i,1] = j + origin[1]

Build network for binocular downstream task It takes an extra argument head, that is called with the features and a dictionary img

Add camera parameters to the observation dictionnary produced by Habitat-Sim In-place modifications.