Compute the inverse of each 4x4 (or 3x4) SE3 matrix in a batch. If `R` and `T` are provided, they must correspond to the rotation and transl

Copy data to the specified device

Converts a depth map to world coordinates (HxWx3) given the camera extrinsic and intrinsic. Returns both the world coordinates and the interm

returns (is_good, err_msg)

This function: - first downsizes the image with LANCZOS inteprolation, which is better than bilinear interpolation in

Open an image or a depthmap with opencv-python.

Thresholds a depth map using percentile-based limits and optional maximum depth clamping. Steps: 1. If `max_depth > 0`, clamp all valu

Checks if 'input_tensor' contains inf or nan values and clamps extreme values. Args: input_tensor (torch.Tensor): The loss tenso

Processes specialized tokens with shape (1, 2, X, C) for multi-frame processing: 1) Uses the first position (index=0) for the first frame onl

Check if a tensor contains NaN or Inf values and log a warning if found. Args: input_tensor: The tensor to check name: N

Implementation of the forward pass through the DPT head. This method processes a specific chunk of frames from the sequence.

Process global attention blocks. We keep tokens in shape (B, S*P, C).

Apply basic activation function to pose parameters. Args: pose_enc: Tensor containing encoded pose parameters act_type: Acti

Convert camera extrinsics and intrinsics to a compact pose encoding. This function transforms camera parameters into a unified pose encoding form

Apply a geometric transformation to a list of 3-D points. H: 3x3 or 4x4 projection matrix (typically a Homography) p: numpy/torch/tuple of c

Apply inverse log transform: sign(y) * (exp(|y|) - 1) Args: y: Input tensor Returns: Transformed tensor

Transfer some variables to another device (i.e. GPU, CPU:torch, CPU:numpy). batch: list, tuple, dict of tensors or other things device: pyto

Applies 1D rotary position embeddings along one dimension. Args: tokens: Input token features. positions: Position in

Apply positional embedding to tensor x.

Process frame attention blocks. We keep tokens in shape (B*S, P, C).

Custom interpolate to avoid INT_MAX issues in nn.functional.interpolate.

Args: - depthmap (HxW array): - camera_intrinsics: a 3x3 matrix - camera_pose: a 4x3 or 4x4 cam2world matrix Returns:

Downloads a file from a Hugging Face model repo, handling redirects.

Filter loss tensor by keeping only values below a certain quantile threshold. This helps remove outliers that could destabilize training

this will perform the index select, cat the tensors, and provide the attn_bias from cache