AstroVis: A Cloud-Streaming Octree System for Web-Based Astronomical Visualization

Large sky surveys and high-resolution planetary products are now routinely distributed as massive point catalogs and tiled textures; yet, many interactive viewers still require heavyweight installations, high-end GPUs, or locally preprocessed data. We present AstroVis, a browser-based system that enables exploratory visualization of large-scale, volumetric star catalogs and globe browsing, while maintaining a lightweight, portable client. AstroVis preprocesses user-uploaded catalogs on a server into a cloud-resident octree, serializes leaf nodes as binary chunks, and streams only visible nodes to the browser using view-frustum culling and a memory-bounded cache. We evaluate AstroVis on two real catalogs (a 3M-star Gaia DR2 subset, 1.1GB; and a 700K-star LAMOST subset, 500MB) and a globe dataset over a 10 Mb/s wide-area link. Upload and octree construction take 8 and 3 minutes, respectively, and per-node transfers average 9.6 seconds. On an integrated GPU, AstroVis sustains interactive rendering comparable to OpenSpace while keeping memory usage manageable. These results suggest a practical path to portable, web-first access to large astronomical datasets for both research and public engagement.