Save RGB thumbnails to cache as a compressed npz archive.

Save sampling metadata to sampling.json.

Save scene-level cross-method comparison entry.

Save per-frame binary sky masks (True = sky pixel) to cache.

Save point cloud animation as video.

Scan workspace and return all valid BSS entries plus a pre-built structure. Validity is determined by the presence of a .complete.json file in th

Forward pass through the fusion blocks. Args: features (List[Tensor]): List of feature maps from different layers.

Segments sky from an image using an ONNX model. Args: image_path: Path to input image onnx_session: ONNX runtime session wit

Add colored border to image.

Redirect stdout/stderr to /dev/null for the process lifetime.

Convert a unit quaternion to a standard form: one in which the real part is non negative. Args: quaternions: Quaternions with re

Convert a unit quaternion to a standard form: one in which the real part is non negative. Args: quaternions: Quaternions with re

Return metrics without rError/tError, for Layer 2 comparison view.

Print global FPS (total time / total frames) + 10/50/90% windows + trace.

Calculate translation angle error between ground truth and predicted translations. Args: tvec_gt: Ground truth translation vectors

Iteratively refine camera pose predictions. Args: pose_tokens (torch.Tensor): Normalized camera tokens with shape [B, 1,

Iteratively refine camera pose predictions. Args: pose_tokens (torch.Tensor): Normalized camera tokens with shape [B, S,

ts tx ty tz qx qy qz qw' → 4×4 T_world_base.

Show all frames up to the current timestep (or only the current one in 4D mode).

Convert depth map to pseudo-color visualization. Args: depth: HxW depth map (meters) min_depth: Minimum depth value for normaliza

Add camera frames and frustums to the scene.

Write camera intrinsics file in BSS v2 format. Args: intr_file: Output file path intrinsics: np.ndarray of shape (N, 4) or List[

Write sampling.json file with frame IDs and metadata. Args: sampling_file: Output file path frame_ids: List of sampled frame IDs

Write trajectory in BSS v2 format. Args: traj_file: Output file path poses: Either np.ndarray of shape (N, 4, 4) or List[Opti

Zoom by moving eye along view direction.

Generates spatial positions for a batch of patches. Args: batch_size: Number of samples in the batch. height: Height

Initialize cache manager. Args: data_dir: Path to method directory (e.g., workspace/dataset/scene/method) spatial_sub

Initialize dataset. Args: raw_data_root: Path to raw dataset directory logger: Optional logger instance

Initialize resize context. Args: mode: Resize strategy. One of: 'none', 'area_budget' align: Alignment r

Initialize BSS saver. Args: artifact: BSSArtifact describing the output directory context: Optional dataset or met

Initialize configuration manager. Args: config_path: Path to base YAML configuration file.

Initialize BSS loader. Args: artifact: BSSArtifact describing the directory to load from resize_context: Option

Initialize evaluator. Args: gt_artifact: BSSArtifact for ground truth pred_artifact: BSSArtifact for method output

Initialize method. Args: area_budget: Maximum pixel area for 'area_budget' resize mode. When provided,

Initialize depth evaluator. Args: align: Alignment method - 'lstsq', 'scale_only', or 'none' gt_clip: (min, max

Initialize point cloud evaluator. Args: dataset: Dataset instance with optional evaluate_pointcloud() method options:

Initialize AUC evaluator. Args: thresholds: List of angle thresholds in degrees (default: [3, 5, 15, 30])

Initialize trajectory evaluator. Args: align: Whether to align trajectories before evaluation correct_scale: Whether

Initialize 7Scenes dataset loader. Args: raw_data_root: Dataset root directory split: 'train' or 'test' (default: 'te

Initialize VBR loader. Args: raw_data_root: Root containing `{scene}_processed_aligned/` dirs and a si

Initialize KITTI Odometry loader. Args: raw_data_root: Dataset root containing `poses/` and `sequences/`. sequences:

Initialize general dataset loader. Args: raw_data_root: Path to an image directory OR a video file. load_img_size: Ta

Initialize Neural RGB-D dataset loader. Args: raw_data_root: Path to dataset root (contains per-scene subdirectories).

Initialize Tanks and Temples dataset loader. Args: raw_data_root: Dataset root directory (containing scene subdirectories)

Initialize the TUM RGB-D dataset loader. Args: raw_data_root: Dataset root directory (e.g., /data0/clz/TUM-RGBD) load

Initialize DROID-W dataset loader. Args: raw_data_root: Root directory containing downtown*/ subdirs. scenes:

Initialize AggregatorStream. Args: sliding_window_size: Sliding window size in blocks (-1 for full causal) n

Initializes the 2D RoPE module.

Args: img_size (int, tuple): input image size patch_size (int, tuple): patch size in_chans (int): number

Init. Args: features (int): number of features

Init. Args: features (int): number of features

Init. Args: features (int): number of features

Initialize GCTStream. Args: img_size: Input image size patch_size: Patch size for embedding embe

Initialize GCTStream. Args: img_size: Input image size patch_size: Patch size for embedding embe

Initialize GCTStream. Args: img_size: Input image size patch_size: Patch size for embedding embe

Args: keyframes: List of dicts with keys: frame, eye, center, up, fov. Must be sorted by frame and cover f