How to Configure YAML Presets in LingBot-Map for Indoor vs Outdoor Scenes

Configure YAML presets in LingBot-Map by passing the --config flag to rendering scripts like demo_render/batch_demo.py, selecting demo_render/config/indoor.yaml for small-scale environments with shallow depth or demo_render/config/outdoor_drive.yaml for large-scale scenes with sky masking and multi-segment camera trajectories.

LingBot-Map (Robbant/lingbot-map) drives its rendering and preprocessing pipelines through declarative YAML preset files. These configuration files act as the source of truth for voxelization parameters, camera paths, and post-processing effects, allowing you to switch between indoor walkthroughs and outdoor drives without modifying Python code.

YAML Preset Structure and Schema

Every preset follows a hierarchical schema where top-level keys group related functionality. When you launch a rendering job, the parser loads the YAML into a configuration object; any CLI flag matching a top-level key will override that specific value, while nested structures like camera.segments remain intact unless explicitly replaced.

The canonical schema includes these sections:

• scene – Voxelization and point-cloud parameters (voxel_size, octree_level, max_depth, downsample)
• preprocess – Per-frame filtering (mask_sky, sky_model, conf_threshold)
• camera – Virtual camera trajectory (fov, transition, segments list)
• render – Engine settings (width, height, point_size, depth_colormap)
• overlay – UI elements for the final video (camera_vis, trail_color_ramp)
• pipeline – Execution control (fps, num_workers, fast_review)
• gpu – Memory tuning (build_batch_size, cull_chunk_size, memory_limit_gb)

Indoor Scene Configuration

For small-scale indoor walkthroughs such as office tours or museum captures, use demo_render/config/indoor.yaml. This preset optimizes for confined spaces and static lighting.

Key settings in the indoor preset:

• scene.max_depth: 10.0 – Shallow rendering depth prevents background noise in tight corridors
• preprocess.mask_sky: false – Disables sky segmentation since indoor scenes lack open sky
• camera.segments – Single birdeye segment covering the entire sequence (frames: [0, -1]) for a stable top-down overview

The shallow depth and disabled sky masking reduce GPU memory consumption and processing time for high-resolution indoor scans.

Outdoor Scene Configuration

For large-scale outdoor drives or aerial footage, use demo_render/config/outdoor_drive.yaml. This preset accommodates vast distances and dynamic lighting conditions.

Key settings in the outdoor preset:

• scene.max_depth: 250.0 – Deep rendering depth captures distant landscape features
• preprocess.mask_sky: true – Enables sky masking using an ONNX segmentation model to exclude overexposed pixels from the point cloud
• camera.segments – Two distinct phases: a follow segment for the first 3000 frames that simulates vehicle-like motion with back_offset, up_offset, and smooth_window parameters, followed by a birdeye segment for the final overview

The sky-segmentation model runs on GPU with a configurable sky_batch_size to balance throughput against memory usage.

Applying Presets via the Command Line

Pass presets to the rendering pipeline using the --config argument. Both demo_render/batch_demo.py and demo.py accept this flag.

For a fully YAML-driven indoor render, avoid passing camera-related flags that would discard the preset's segments list:

python demo_render/batch_demo.py \
  --video_path /data/indoor.mp4 \
  --output_folder /data/out/indoor \
  --model_path /path/to/lingbot-map.pt \
  --config demo_render/config/indoor.yaml \
  --mode windowed --window_size 128 \
  --keyframe_interval 10

For outdoor scenes, you can keep the preset and override specific values on the fly. For example, increasing camera optimization iterations without editing the file:

python demo_render/batch_demo.py \
  --video_path /data/drive.mp4 \
  --output_folder /data/out/drive \
  --model_path /path/to/lingbot-map.pt \
  --config demo_render/config/outdoor_drive.yaml \
  --camera_num_iterations 6

Creating Custom Presets

To customize behavior, copy an existing preset and modify the fields. The rendering pipeline automatically picks up new values without code changes.

For example, create demo_render/config/custom_indoor.yaml with a longer chase-camera segment:

scene:
  voxel_size: 0.001
  max_depth: 12.0
preprocess:
  mask_sky: false
camera:
  fov: 60.0
  transition: 30
  segments:
    - mode: follow
      frames: [0, 5000]
      back_offset: 0.25
      up_offset: 0.07
      look_offset: 0.35
      smooth_window: 25
    - mode: birdeye
      frames: [5000, -1]
      reveal_height_mult: 2.0

Run the custom preset:

python demo_render/batch_demo.py \
  --video_path /data/custom_indoor.mp4 \
  --output_folder /data/out/custom_indoor \
  --model_path /path/to/lingbot-map.pt \
  --config demo_render/config/custom_indoor.yaml

Summary

• LingBot-Map uses YAML presets in demo_render/config/ to declaratively control rendering, preprocessing, and camera behavior
• Indoor presets (indoor.yaml) use shallow depth (max_depth: 10.0), disable sky masking, and employ a single birdeye camera segment
• Outdoor presets (outdoor_drive.yaml) use deep depth (max_depth: 250.0), enable ONNX-based sky segmentation, and combine follow and birdeye camera segments
• Pass presets via --config to batch_demo.py or demo.py; CLI flags override top-level YAML keys but avoid camera flags when relying on preset segments
• Copy and modify existing YAML files to create custom configurations without altering source code

Frequently Asked Questions

What happens if I pass camera flags while using a YAML preset?

If you pass flags like --camera_mode, --back_offset, or --look_offset on the command line, the renderer will discard the camera.segments list defined in the YAML and use the CLI values instead. To preserve the preset's multi-segment trajectory, avoid passing any camera-related flags and control the behavior solely through the YAML file.

Can I switch between indoor and outdoor presets without reprocessing the point cloud?

Yes. The YAML presets primarily affect the rendering and camera trajectory phases. If you have already generated a point cloud, you can re-run the rendering stage with a different --config file to produce new videos with different camera paths or depth settings without recomputing the underlying scene geometry.

How do I adjust GPU memory limits for large outdoor scenes?

Modify the gpu section in your YAML preset. Increase memory_limit_gb to allow larger working sets, or decrease build_batch_size and cull_chunk_size to process the scene in smaller chunks. These settings are located under the gpu: key in both indoor.yaml and outdoor_drive.yaml.

Why does the outdoor preset require an ONNX model?

The outdoor preset sets preprocess.mask_sky: true to exclude sky pixels from the point cloud, preventing floating artifacts at the horizon. This feature requires an ONNX-format segmentation model specified by the sky_model path. The model runs inference on each frame to generate binary sky masks before depth fusion occurs.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}