### Install and Download OmniWorld Dataset

Source: https://github.com/yangzhou24/omniworld/blob/main/README.md

Installs the huggingface_hub library and downloads the full OmniWorld dataset to a specified local directory. Ensure you have the CLI installed.

```bash
# 1. Install (if you haven't yet)
pip install --upgrade "huggingface_hub[cli]"

# 2. Full download
hf download InternRobotics/OmniWorld \
           --repo-type dataset \
           --local-dir /path/to/DATA_PATH
```

--------------------------------

### Install and Download Full OmniWorld Dataset

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Installs the Hugging Face CLI and downloads the entire OmniWorld dataset to a specified local directory.

```bash
pip install --upgrade "huggingface_hub[cli]"

hf download InternRobotics/OmniWorld \
           --repo-type dataset \
           --local-dir /path/to/DATA_PATH
```

--------------------------------

### Camera JSON Format Example

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Defines the JSON format for camera pose files, including focal lengths, principal points (cx, cy), quaternion rotations, and translation vectors.

```json
{
  "focals": [525.0, 525.0, 525.0],
  "cx": [320.0, 320.0, 320.0],
  "cy": [240.0, 240.0, 240.0],
  "quats": [
    [1.0, 0.0, 0.0, 0.0],
    [0.999, 0.01, 0.0, 0.0],
    [0.998, 0.02, 0.0, 0.0]
  ],
  "trans": [
    [0.0, 0.0, 0.0],
    [0.1, 0.0, 0.0],
    [0.2, 0.0, 0.0]
  ]
}
```

--------------------------------

### Visualize Scene as Point Cloud (Metric Scale)

Source: https://github.com/yangzhou24/omniworld/blob/main/README.md

Converts a scene from the OmniWorld-Game dataset into a metric-scaled 3D point cloud using the provided script and metadata. The output is saved as a .ply file.

```python
python scripts/visualize_pcd.py <your-data-path>/b04f88d1f85a --split_idx 0 --metadata_csv metadata/omniworld_game_metadata.csv
```

--------------------------------

### Project Depth Maps to 3D Coordinates

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Projects depth maps to 3D world coordinates using camera intrinsics and extrinsics. Returns global 3D point positions for each pixel. Requires loading camera poses and depth maps.

```python
import numpy as np
from scripts.utils import project, load_camera_poses, load_depth

# Load camera data
scene_dir = Path("/path/to/scene_data/b04f88d1f85a")
intrinsics, extrinsics, frame_indices = load_camera_poses(scene_dir, split_idx=0)

# Load depth maps for each frame
depths = []
for idx in frame_indices:
    depth, _ = load_depth(f"{scene_dir}/depth/{idx:06d}.png")
    depths.append(depth)
depths = np.stack(depths, axis=0)  # (N, H, W)

# Project to 3D world coordinates
points = project(intrinsics, extrinsics, depths)
print(f"Points shape: {points.shape}")  # (N, H, W, 3) - 3D coords for each pixel
```

--------------------------------

### Load Camera Poses (Intrinsics and Extrinsics)

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Loads camera intrinsics and extrinsics for a specific split within a scene directory. Optionally rescales poses to metric scale using provided metadata.

```python
from pathlib import Path
from scripts.utils import load_camera_poses

scene_dir = Path("/path/to/scene_data/b04f88d1f85a")
split_idx = 0
metric_scale = 1.25  # Optional: from metadata CSV

# Load camera poses
intrinsics, extrinsics, frame_indices = load_camera_poses(
    scene_dir,
    split_idx=split_idx,
    metric_scale=metric_scale
)

print(f"Loaded {len(frame_indices)} frames")
print(f"Intrinsics shape: {intrinsics.shape}")  # (S, 3, 3) pixel-space K matrices
print(f"Extrinsics shape: {extrinsics.shape}")  # (S, 4, 4) OpenCV world-to-camera matrices
print(f"Frame indices: {frame_indices}")
```

--------------------------------

### Visualize Scene as Point Cloud (Relative Scale)

Source: https://github.com/yangzhou24/omniworld/blob/main/README.md

Converts a scene from the OmniWorld-Game dataset into a relative-scaled 3D point cloud using the provided script. This version does not use external metadata for scaling.

```python
python scripts/visualize_pcd.py <your-data-path>/b04f88d1f85a --split_idx 0
```

--------------------------------

### Generate Point Cloud from Scene Data

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Converts a scene's depth maps and camera poses into a 3D point cloud. Supports generation with relative scale or metric scale using metadata.

```bash
# Generate point cloud with relative scale (no metadata)
python scripts/visualize_pcd.py /path/to/scene_data/b04f88d1f85a --split_idx 0
# Output: split0_points.ply

# Generate point cloud with metric scale (real-world units)
python scripts/visualize_pcd.py /path/to/scene_data/b04f88d1f85a \
    --split_idx 0 \
    --metadata_csv metadata/omniworld_game_metadata.csv
# Output: split0_points_metric.ply
```

--------------------------------

### Load Depth Map with Validity Mask

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Loads a 16-bit depth map from a PNG file, returning depth values and a validity mask. Optionally converts depth to metric scale.

```python
import os
from scripts.utils import load_depth

depth_path = "/path/to/scene_data/b04f88d1f85a/depth/000001.png"
metric_scale = 1.25  # Optional: from metadata CSV

# Load depth map
depthmap, valid_mask = load_depth(depth_path, metric_scale=metric_scale)

print(f"Depth shape: {depthmap.shape}")      # (H, W) float32
print(f"Valid mask shape: {valid_mask.shape}")  # (H, W) bool
print(f"Valid pixels: {valid_mask.sum()}")
print(f"Depth range (valid): {depthmap[valid_mask].min():.2f} - {depthmap[valid_mask].max():.2f}")
# Invalid pixels (too close, sky, far areas) are set to -1
```

--------------------------------

### OmniWorld Scene Directory Structure

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Illustrates the expected directory structure for a scene within the OmniWorld dataset, detailing the types of annotation files available.

```plaintext
<your-data-path>/b04f88d1f85a/
├── color/              # RGB frames (.png)
├── depth/              # 16-bit depth maps (.png)
├── flow/               # Optical flow: flow_u_16.png, flow_v_16.png, flow_vis.png
├── camera/             # split_*.json (intrinsics + extrinsics per split)
├── subject_masks/      # Foreground masks (per split)
├── gdino_mask/         # Dynamic-object masks (per frame)
├── text/               # Structured captions (81-frame segments)
├── droidclib/          # Coarse camera poses (optional)
├── fps.txt             # Source video framerate
└── split_info.json     # Frame groupings into splits
```

--------------------------------

### Load Metric Scale from CSV

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Loads the per-scene metric scale factor from a metadata CSV file. Requires scene directory and metadata CSV path. Handles KeyError if the scene is not found.

```python
from pathlib import Path
from scripts.utils import load_metric_scale

scene_dir = Path("/path/to/scene_data/b04f88d1f85a")
metadata_csv = Path("metadata/omniworld_game_metadata.csv")

try:
    metric_scale = load_metric_scale(scene_dir, metadata_csv)
    print(f"Metric scale for {scene_dir.name}: {metric_scale}")
except KeyError as e:
    print(f"Scene not found in metadata: {e}")
```

--------------------------------

### Extract Scene Data (RGB, Depth, Flow, Annotations)

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Extracts RGB frames, depth maps, optical flow, and other annotations from downloaded tar archives for a specific scene ID. Ensure the root path points to your OmniWorld data directory.

```bash
scene_id=1e527547f893
root=/path/to/DATA_PATH        # where you store OmniWorld

mkdir -p ${scene_id}

# Extract RGB frames (may span multiple parts)
for rgb_tar in ${root}/videos/OmniWorld-Game/${scene_id}/${scene_id}_rgb_*.tar.gz
do
    echo "Extracting $(basename $rgb_tar)…"
    tar -xzf "$rgb_tar" -C ${scene_id}
done

# Extract depth maps
for d_tar in ${root}/annotations/OmniWorld-Game/${scene_id}/${scene_id}_depth_*.tar.gz
do
    echo "Extracting $(basename $d_tar)…"
    tar -xzf "$d_tar" -C ${scene_id}
done

# Extract optical flow
for f_tar in ${root}/annotations/OmniWorld-Game/${scene_id}/${scene_id}_flow_*.tar.gz
do
    echo "Extracting $(basename $f_tar)…"
    tar -xzf "$f_tar" -C ${scene_id}
done

# Extract all other annotations (camera poses, masks, text captions)
tar -xzf ${root}/annotations/OmniWorld-Game/${scene_id}/${scene_id}_others.tar.gz -C ${scene_id}
```

--------------------------------

### Download Specific Scenes from OmniWorld-Game

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Downloads specific scenes from the OmniWorld-Game subset using either a metadata CSV file or by directly providing scene UIDs. Supports authentication for private repositories.

```python
# Download specific rows from metadata CSV
python scripts/download_specific.py \
    --csv omniworld_game_metadata.csv \
    --repo InternRobotics/OmniWorld \
    --local_dir ./OmniWorld_subset \
    --rows 0 1

# Or specify UIDs directly (bypassing CSV)
python scripts/download_specific.py \
    --uids 0b3caf48b79d 0c3cefcf3a15 \
    --local_dir ./OmniWorld_subset

# For private repos, include authentication token
python scripts/download_specific.py \
    --uids 0b3caf48b79d \
    --local_dir ./OmniWorld_subset \
    --token YOUR_HF_TOKEN
```

--------------------------------

### Export 3D Point Cloud to PLY

Source: https://context7.com/yangzhou24/omniworld/llms.txt

Exports 3D point cloud data to a PLY file. Supports custom RGB colors or auto-generated colors based on spatial coordinates. Requires points and optionally RGB colors.

```python
import numpy as np
from scripts.utils import write_ply

# Points: (N, 3) array of XYZ coordinates
points = np.random.randn(10000, 3).astype(np.float32)

# RGB colors: (N, 3) array, values 0-255 or 0.0-1.0
colors = np.random.randint(0, 256, (10000, 3)).astype(np.uint8)

# Write PLY with custom colors
write_ply(points, rgb=colors, path="output_colored.ply")

# Write PLY with auto-generated colors based on position
write_ply(points, rgb=None, path="output_auto_color.ply")
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.