How to Adjust LingBot-Map Visualization Options: conf_threshold, point_size, and downsample_factor

Adjust LingBot-Map visualization options like conf_threshold, point_size, and downsample_factor either via command-line arguments in demo.py, through interactive GUI sliders at runtime, or programmatically when instantiating the PointCloudViewer class.

LingBot-Map provides flexible control over 3D point cloud rendering through the PointCloudViewer class located in lingbot_map/vis/point_cloud_viewer.py. These visualization parameters allow you to filter low-confidence points, adjust the apparent size of rendered points, and control spatial downsampling density to balance visual detail against GPU performance. Whether you are running the interactive demo or embedding the viewer in your own pipeline, you can configure these settings statically or change them dynamically without restarting the application.

Command-Line Configuration for Static Settings

When launching the interactive demo via demo.py, you can set initial values for all three visualization options using CLI flags. The parser accepts --conf_threshold, --point_size, and --downsample_factor at lines 406–408, forwarding them to the PointCloudViewer constructor.

The available flags and their default values are:

• --conf_threshold (default: 1.5): Sets the minimum confidence score (vis_threshold) required for a point to be displayed.
• --point_size (default: 0.00001): Controls the base radius of each rendered point in the 3D scene.
• --downsample_factor (default: 10): Determines the spatial downsampling rate applied before rendering; lower values preserve more points and increase detail.

Example launch command with custom visualization settings:

python demo.py \
    --model_path /path/to/lingbot-map.pt \
    --image_folder example/courthouse \
    --conf_threshold 2.0 \
    --point_size 0.00005 \
    --downsample_factor 5

Interactive GUI Adjustment at Runtime

The PointCloudViewer exposes interactive sliders that map directly to these parameters, allowing real-time tweaking without restarting the server. These controls are initialized in the _setup_gui() method around line 400 in lingbot_map/vis/point_cloud_viewer.py.

To adjust settings dynamically:

1. Launch the viewer and navigate to http://localhost:8080.
2. Locate the GUI panel containing three sliders:
   • Visibility Threshold: Adjusts the confidence cutoff (vis_threshold). Slide right to filter out low-confidence points.
   • Point Size: Modifies the render size (point_size). Slide right to increase point diameter.
   • Downsample Factor: Controls sampling density (downsample_factor). Slide left to reduce downsampling and increase point density.

When you move a slider, its on_update callback triggers _regenerate_point_clouds() (around line 730), which immediately re-filters and re-renders the point cloud using the new values.

Programmatic Configuration in Python Scripts

For custom applications, instantiate PointCloudViewer directly and pass the desired values to its constructor. This bypasses the CLI parser in demo.py and allows hardcoded or dynamically computed settings.

from lingbot_map.vis import PointCloudViewer

viewer = PointCloudViewer(
    pred_dict=predictions,
    port=8080,
    vis_threshold=2.0,      # Maps to conf_threshold

    point_size=0.00005,     # Point radius

    downsample_factor=5,    # Lower values retain more points

)
viewer.run()

In demo.py (lines 84–90), this is precisely how the viewer is initialized, mapping args.conf_threshold to the vis_threshold parameter, args.downsample_factor to downsample_factor, and args.point_size to point_size.

Internal Implementation and Data Flow

Understanding the internal pipeline helps optimize these settings for your specific hardware and data quality.

The visualization pipeline operates in three stages inside point_cloud_viewer.py:

1. Initialization: The constructor stores vis_threshold, point_size, and downsample_factor as instance attributes.

2. Data Parsing: The parse_pc_data() method (around line 660) applies the filters:

   • Removes points with confidence scores below self.vis_threshold.
   • Applies voxel downsampling using self.downsample_factor.
   • Passes the resulting point cloud to the renderer with point_size=self.psize_slider.value.

3. Regeneration: When sliders change, _regenerate_point_clouds() re-executes the parsing logic and updates the scene handles instantly.

This architecture ensures that heavy computation occurs only when parameters change, maintaining interactive frame rates during camera manipulation.

Summary

• Command-line flags (--conf_threshold, --point_size, --downsample_factor) in demo.py set static initial values parsed at lines 406–408.
• Interactive sliders in the web GUI provide real-time control over confidence filtering, point size, and downsampling density without restarting the server.
• Programmatic access allows direct instantiation of PointCloudViewer with custom vis_threshold, point_size, and downsample_factor arguments.
• All changes trigger _regenerate_point_clouds() in lingbot_map/vis/point_cloud_viewer.py to update the rendered output immediately.

Frequently Asked Questions

What is the default conf_threshold in LingBot-Map?

The default confidence threshold is 1.5, set in demo.py and passed to the PointCloudViewer as vis_threshold. Points with confidence scores below this value are filtered out during the initial render and remain hidden until the threshold is lowered via the CLI or the "Visibility Threshold" slider.

How does downsample_factor affect performance and quality?

The downsample_factor controls spatial voxel downsampling before rendering. A value of 10 (the default) means aggressive downsampling for higher performance, while setting it to 1 retains nearly all points for maximum detail at the cost of GPU memory and frame rate. Adjust this based on your point cloud density and graphics hardware capabilities.

Can I change visualization settings without restarting the viewer?

Yes. The PointCloudViewer provides interactive sliders for Visibility Threshold, Point Size, and Downsample Factor in the browser interface. These sliders update the underlying attributes and call _regenerate_point_clouds() immediately, allowing you to tune the visualization in real-time without reloading the application or reprocessing the source images.

Where are the GUI sliders defined in the source code?

The sliders are created in the _setup_gui() method of lingbot_map/vis/point_cloud_viewer.py around line 400. Each slider (self.vis_threshold_slider, self.psize_slider, self.downsample_slider) registers an on_update callback that synchronizes the GUI state with the renderer and triggers point cloud regeneration.

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