what is flux-gs fr?

Flux-GS is the research code for a paper called Monte Carlo Energy Aggregation for Mobile 3D Gaussian Splatting, accepted at the European Conference on Computer Vision. Gaussian Splatting is a technique for representing a 3D scene as a large collection of small blobs of color and opacity, which can be rendered very quickly compared to older methods, and this project focuses specifically on making that technique fast enough and small enough to run on mobile devices in a web browser. The README states that the code lets you train these compressed Gaussian Splatting scenes and then render them on mobile through a WebGL renderer, which is released as a separate, linked repository. It builds on the authors' earlier work called Mobile-GS, along with two other referenced projects, MVGS and FastGS, and the trained scene file is saved as a JSON format designed to load efficiently in a browser using point cloud and a small neural network for decompression. To use it you need a machine with an NVIDIA GPU running CUDA 12.6 and Python 3.11, PyTorch, and a separate compression tool called TMC that has to be installed and configured manually. Training and evaluation are done with standard benchmark datasets used in this research area, such as Mip-NeRF 360 and Tanks and Temples, and the authors also released pretrained weights on Hugging Face so you can run evaluation without training from scratch. A note in the README explicitly states the ideas in Flux-GS are free to use for commercial purposes.