Streaming Point Cloud Visualizer
City-scale LiDAR scans are too large to load into GPU memory. This viewer makes them feel local: users can smoothly fly through an entire city while detail streams in around the camera.
The Vulkan renderer displays a 180 GB, 13-billion-point scan of San Francisco at 120 FPS with a 4 GB GPU budget. The full data set never needs to fit in device memory.
Adaptive level of detail prioritizes point density in screen space, while asynchronous, chunked loading moves the relevant data from disk to GPU as the viewpoint changes.
A custom compute-shader rasterizer uses a visibility buffer and 64-bit atomics to resolve the visible surface efficiently, enabling smooth navigation through billions of points.