WilLaGS: Robust Gaussian Splatting in-the-Wild via Latent Appearance Fields

zhenjun_zhao · x · 2026-08-31

The paper "WilLaGS" proposes a unified framework for robust 3D scene reconstruction and generative appearance synthesis in unconstrained settings. It introduces a generative appearance model using a β-VAE to learn a continuous manifold of global appearance and constructs a 3D neural appearance field for dynamic Tri-Plane features. To handle transient objects, it uses a self-supervised perceptual masking mechanism with a Teacher-Student (EMA) architecture. Experiments show state-of-the-art performance in reconstruction and novel view synthesis.

Original post →

More from Research

Research channel →