01
Monte Carlo energy aggregation
Samples third-order radiance residuals and aggregates specular energy into a compact latent space for efficient lower-order representation.
Specular energy ECCV 2026
Flux-GS
Monte Carlo Energy Aggregation for Mobile 3D Gaussian Splatting
Monte Carlo energy aggregation for high-fidelity, efficient 3D Gaussian Splatting on mobile platforms.
Real-time mobile capture
Fluent, free-viewpoint rendering of challenging unbounded scenes—directly on a phone.
01 / Overview
Desktop-class rendering.
Now in your pocket.
High-order spherical harmonics preserve rich view-dependent appearance, but their inference and storage overhead make them difficult to deploy on resource-constrained mobile platforms.
Flux-GS introduces a Monte Carlo Specular Energy Aggregator that samples third-order radiance residuals and gathers visually salient specular energy into a compact latent space. This preserves expressive lighting in lower-order bands without distillation or pre-training.
An Attribute-Conditioned SH Enhancement module recovers high-frequency detail through Gaussian-aware offsets before inference. Multi-view alpha guidance further improves structural consistency while removing redundant primitives.
Live WebGL demo
Try it on
your phone.
No install. No special app. Choose a scene and experience real-time Gaussian Splatting directly in your mobile browser. Access to Hugging Face is required to load scene data.
Pick a scene02 / Method
Preserve energy.
Reduce overhead.
Three complementary components retain expressive appearance while building a compact, mobile-ready representation.
02
Attribute-conditioned SH enhancement
Predicts Gaussian-aware offsets from intrinsic attributes to enhance first-order spherical harmonics before inference.
No inference overhead03
Multi-view alpha guidance
Guides densification and pruning across multiple views to improve structural consistency and precisely remove redundant Gaussians.
Consistent & compact