Urban-GS Viewer & Renderer

A lightweight web viewer and offline renderer for Urban-GS: A Unified 3D Gaussian Splatting Framework for Compact and High-Fidelity Aerial-to-Street Reconstruction models, built on top of nerfstudio.

The core codebase is derived from nerfstudio. The Gaussian-based rendering framework also follows design ideas from yzslab/nerfstudio. We appreciate the open-source work and contributions from both projects.

Visualization_example.mp4

Note: The web viewer preview may exhibit lower visual quality than the actual rendered output video.

Notes

• The camera control logic has been updated to align with SIBR-style navigation. Use WASDQE and IJKLUO to control the camera.
• Dynamic blur control has been disabled so the rendering resolution stays stable while the camera is moving.
• This project is focused on rendering only. Training is not supported, the implementation has not been extensively tested, and it is not as fast as the SIBR viewers.
• The nerfstudio web viewer can reach similar rendering quality to SIBR viewers. Recommended settings:
  • Increase the render resolution with the Max Res option under the CONTROLS tab.
  • Increase JPEG quality with --config.viewer.jpeg-quality 100, or switch to PNG output with --config.viewer.image-format png.
• The default scene orientation may be incorrect. You can adjust it in one of the following ways:
  • Add --no-auto-reorient to run_viewer.py or render.py if you want to keep the same coordinate system as the input dataset.
  • Click RESET UP DIRECTION under the SCENE tab to use the current viewpoint as the orientation.
  • Use --ref-orientation IMAGE_NAME to specify an image as the reference orientation.
• Hold the right mouse button and move the camera slightly before pressing W; otherwise, the camera may freeze after moving a short distance.
  • A fix has been submitted upstream: nerfstudio-project/nerfstudio#2404
  • Alternatively, use this viewer: https://nsv.cslab.pro/23-09-15-1/?websocket_url=ws://localhost:7007
• Press F5 to refresh the page if the web viewer stops updating. If you are creating a camera path, click EXPORT PATH to download it before refreshing, then use LOAD PATH to restore it afterward.

Installation

1. Install the nerfstudio dependencies.
2. Run pip install -e . in this repository. This is required if you are reusing a virtual environment from another nerfstudio checkout.
3. Install the additional dependencies:

pip install plyfile==0.8.1
pip install ./submodules/gsplat-urbangs
pip install ./submodules/simple-knn

4. Install the viewer frontend dependencies:

cd nerfstudio/viewer/app
npm install
# or
yarn install

Usage

Viewer

Start the viewer frontend:

cd nerfstudio/viewer/app
yarn start

In a separate terminal, launch the Urban-GS viewer backend:

python nerfstudio/scripts/urban_gs/run_viewer.py --model-path GAUSSIAN_TRAINING_OUTPUT_MODEL_DIR

Render

Render a camera path to a video:

python nerfstudio/scripts/urban_gs/render.py camera-path \
    --model-path GAUSSIAN_TRAINING_OUTPUT_MODEL_DIR \
    --camera-path-filename YOUR_CAMERA_PATH_FILE.json \
    --output-path YOUR_OUTPUT_MP4_FILE.mp4

License

Code derived from nerfstudio is licensed under the Apache-2.0 license.