Constructs and returns the OpenGL conversion matrix. Returns: numpy.ndarray: A 4x4 OpenGL conversion matrix.

Generate 3D world points from depth predictions. Args: predictions (dict): Prediction dictionary containing pose_enc, depth, and ima

Multi-scale gradient loss wrapper. Applies gradient loss at multiple scales by subsampling the input. This helps capture both fine and coarse

This function generates a 1D positional embedding from a given grid using sine and cosine functions. Args: - embed_dim: The embedding di

depth: [B, V, H, W, 1] mask: [B, V, H, W]

Convert 3D point map to surface normal vectors using cross products. Computes normals by taking cross products of neighboring point diff

Convert a pose encoding back to camera extrinsics and intrinsics. This function performs the inverse operation of extri_intri_to_pose_encoding,

Converts VGGT predictions to a 3D scene represented as a GLB file. Args: predictions (dict): Dictionary containing model predictions

Core regression loss function with confidence weighting and optional gradient loss. Computes: 1. gamma * ||pred - gt||^2 * conf - al

Segments sky from an image using an ONNX model. Thanks for the great model provided by https://github.com/xiongzhu666/Sky-Segmentation-and-Po

Remove batch dimension from all tensors in the inputs dictionary and optionally change data type. Args: inputs (dict): Dictionar

This function disables printing when not in master process

Applies a 4x4 transformation to a set of points. Args: transformation (np.ndarray): Transformation matrix. points (np.ndarra

Unproject a batch of depth maps to 3D world coordinates. Args: depth_map (np.ndarray): Batch of depth maps of shape (S, H, W, 1) or

Update the point cloud based on current GUI selections.

Build the patch embed layer. If 'conv', we use a simple PatchEmbed conv layer. Otherwise, we use a vision transformer.

Initializes the position generator with an empty cache.

Build camera intrinsics matrix from parameters.

Computes frequency components for rotary embeddings. Args: dim: Feature dimension (must be even). seq_len: Maximum se

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

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

Advanced string replacement. Given a text: - replace all elements in src by the corresponding element in tgt - remove all elements in rm

Performs feature rotation by splitting and recombining feature dimensions. Args: x: Input tensor to rotate. Returns:

Returns torch.sqrt(torch.max(0, x)) but with a zero subgradient where x is 0.

Process network output to extract 3D points and confidence values. Args: out: Network output tensor (B, C, H, W) activation:

Activate pose parameters with specified activation functions. Args: pred_pose_enc: Tensor containing encoded pose parameters [transl

Aligns the 3D scene based on the extrinsics of the first camera. Args: scene_3d (trimesh.Scene): The 3D scene to be aligned.

Apply sky segmentation to confidence scores. Args: conf (np.ndarray): Confidence scores with shape (S, H, W) image_folder (s

Build a learning rate scheduler with linear warmup and cosine decay. Args: optimizer: Optimizer instance warmup_steps: N

Build data loader for training or testing. Args: dataset: Dataset configuration string batch_size: Batch size nu

Build multi-task loss criterion. Args: cfg: Configuration object Returns: MultitaskLoss instance

Build optimizer with parameter groups for different learning rates. Args: model: Model instance cfg: Configuration objec

Computes translation, rotation, and focal loss for a batch of pose encodings. Args: pred_pose_enc: (N, D) predicted pose encodin

Modify camera intrinsics to follow a different convention. Coordinates of the center of the top-left pixels are by default: - (0.5, 0.5)

Computes the faces for the camera mesh. Args: cone_shape (trimesh.Trimesh): The shape of the camera cone. Returns: np.n

Compute depth loss. Args: predictions: Dict containing 'depth' and 'depth_conf' batch: Dict containing ground truth 'dep

Compute point loss. Args: predictions: Dict containing 'world_points' and 'world_points_conf' batch: Dict containing gro

Create a normalized UV grid of shape (width, height, 2). The grid spans horizontally and vertically according to an aspect ratio, ensuri

Convert a depth map to camera coordinates. Args: depth_map (np.ndarray): Depth map of shape (H, W). intrinsic (np.ndarray):

Unprojects a depth map into camera coordinates, returning (H, W, 3). Args: depth_map (np.ndarray): Depth map of shape (H

Args: - depthmap (HxW array): - camera_intrinsics: a 3x3 matrix Returns: pointmap of absolute coordinates (HxWx3 arra