Clears the 3D viewer, the stored target_dir, and empties the gallery.

Run pose conditioned depth estimation on COLMAP data.

Create the control section with buttons and options. Returns: Tuple of control components

Create a dataset. Requires the file to contain a "create_dataset" function.

Create the depth visualization section. Returns: A tuple of (prev_depth_btn, depth_view_selector, next_depth_btn, depth_

Create a colored depth visualization. Args: depth: Depth array Returns: Colored depth visualization or None

HTML dashboard for monitoring backend status and tasks.

Delete a specific task.

Evaluate fused point cloud against ScanNet++ ground truth mesh. Uses AABB cropping to only evaluate points within GT bounding box.

Evaluate fused point cloud against DTU GT with ObsMask/Plane. Args: scene: Scene identifier fuse_path: Path

Evaluate fused point cloud against 7Scenes ground truth mesh. Args: scene: Scene identifier fuse_path: Path

Evaluate fused point cloud against HiRoom ground truth point cloud. Args: scene: Scene identifier fuse_path:

Evaluate fused point cloud against ETH3D ground truth mesh. Args: scene: Scene identifier fuse_path: Path to

NOT SUPPORTED for DTU-64. DTU-64 is only for pose evaluation, not 3D reconstruction.

Format value or return N/A.

Parse a string of comma-separated triplets: fn:key:value Returns: dict[fn_name] -> dict[key] = parsed_value Example: "f

Args: feats: List of 4 entries, each entry is a tensor like [B, S, T, C] (or the 0th element of tuple/list is that tensor).

xs: - xs[0]: Top branch input - xs[1]: Lateral input (can do residual addition with top branch)

Args: aggregated_tokens_list: List of 4 tensors [B, S, T, C] from transformer. images: [B, S, 3, H, W]

Forward pass through the network. Args: x: Input images (B, N, 3, H, W) extrinsics: Camera extrinsics (B, N,

Forward pass through both branches with metric scaling alignment. Args: x: Input images (B, N, 3, H, W) extr

Applies 2D rotary position embeddings to input tokens. Args: tokens: Input tensor of shape (batch_size, n_heads, n_tokens, dim).

Fuse per-view depths into a point cloud using TSDF fusion. Args: scene: Scene identifier result_path: Path t

Fuse per-view depths into a point cloud and save to PLY. Args: scene: Scene identifier (e.g., "scan114") res

Fuse per-view depths into a point cloud using TSDF fusion. Args: scene: Scene identifier result_path: Path t

Fuse per-view depths into a point cloud using TSDF fusion. Args: scene: Scene identifier result_path: Path t

Fuse per-view depths into a point cloud using TSDF fusion. Pipeline: 1. Load original images (keep original size) 2.

NOT SUPPORTED for DTU-64. DTU-64 is only for pose evaluation, not 3D reconstruction.

Launch Depth Anything 3 Gallery server

Serve gallery static files (GLB, JPG, etc.).

Get manifest for a specific group.

Gallery home page.

Get gallery group list.

embed_dim: output dimension for each position pos: a list of positions to be encoded: size (M,) out: (M, D)

Get path to ground truth point cloud for a scene.

Collect per-view image paths, intrinsics/extrinsics for a scene. Args: scene: Scene identifier (e.g., "scan105")

Get detailed GPU memory information.

Get metric value or None.

Get pose metric with fallback for different naming conventions.

Get backend status with GPU memory information.

Get task status by task ID.

Launch Depth Anything 3 Gradio interactive web application

Handle file uploads and update gallery. Args: input_video: Path to input video file input_images: List of in

Shorthand for a horizontal linear concatenation.

Run camera pose and depth estimation on a single image.

Run camera pose and depth estimation on a directory of images.

List all tasks.

Load a scene from examples directory. Args: scene_name: Name of the scene to load examples_dir: Path to exam

Load and preprocess image with undistortion and cropping. Args: img_path: Path to image file idx: Index of t

Load pretrained weights for a nested model with both main and metric branches. Args: model: Nested model instance main_model

Load pretrained weights for a single model. Args: model: Model instance to load weights into model_path: Path to the pretrai

Manually trigger task cleanup.

Handle measurement on images. Args: processed_data: Processed data dictionary measure_points: List of curren

Handle measurement on images. Args: processed_data: Processed data dictionary measure_points: List of curren

Navigate depth view. Args: processed_data: Processed data dictionary current_selector: Current selector valu

Navigate depth view (direction: -1 for previous, +1 for next). Args: processed_data: Processed data dictionary

Navigate measure view (direction: -1 for previous, +1 for next). Args: processed_data: Processed data dictionary

Reduce numpy array of shape (49, 27, 36, 3072) to 3D via PCA and visualize as (49, 27, 36, 3). - PCA uses torch.pca_lowrank (randomized SVD),

Process image directory

Process COLMAP data

Process video, extract frames