{"slug": "compact-neural-appearance-models-for-efficient-gaussian-splatting", "title": "Compact Neural Appearance Models for Efficient Gaussian Splatting", "summary": "Researchers Florian Hahlbohm, Jorge Condor, Linus Franke, Martin Eisemann and Marcus Magnor published an arXiv paper introducing a compact neural appearance model for 3D Gaussian Splatting that reduces the per-primitive appearance footprint from 192 to 28 bytes compared with third-degree spherical harmonics, accelerates optimization by 1.3×, and improves reconstruction quality. The work integrates spherical harmonics and recent spherical appearance models into one optimized pipeline with fused forward and backward passes in a differentiable CUDA rasterizer, plus a portable WebGL viewer for laptop and mobile GPUs. The authors report that recent spherical models offer the strongest overall quality–efficiency trade-off, while their neural representation is the most compact model evaluated.", "body_md": "## \n            Compact Neural Appearance Models for Efficient Gaussian Splatting\n\narXiv 2026\n\n### **TL;DR: An efficient appearance framework and a compact neural model for 3DGS.**\n\n### Abstract\n\n                Explicit primitive-based radiance fields such as 3D Gaussian Splatting typically model view-dependent\n                appearance using low-order spherical harmonics (SH). Although efficient to evaluate, SH coefficients\n                dominate per-primitive storage and memory traffic, while their band-limited basis restricts angular\n                detail.\n                \n\n                We present a thorough, end-to-end comparison of SH and recent spherical appearance models and\n                introduce an implicit alternative that decodes compact per-primitive latent codes using a tiny shared\n                MLP. We integrate all models into the same optimized pipeline, fusing their forward and backward passes\n                into a differentiable CUDA rasterizer and provide a portable WebGL viewer for laptop and mobile GPUs.\n                \n\n                Our evaluation across reconstruction quality, memory use, and optimization and rendering performance\n                shows that recent spherical models offer the strongest overall quality–efficiency trade-off. Our neural\n                representation is the most compact model evaluated and, compared to third-degree SH, reduces the\n                per-primitive appearance footprint from 192 to 28 bytes, accelerates optimization by 1.3×, while\n                improving reconstruction quality. We further analyze how appearance parametrization shapes optimization,\n                identifying differences in recovered geometry and the tendency of expressive models to absorb non-static\n                scene content. Together, our framework and analysis provide practical guidance for replacing SH beyond\n                what image metrics alone can capture.\n            \n\n### Citation\n\n```\n@misc{hahlbohm2026efficientgaussianappearance,\n  title         = {Compact Neural Appearance Models for Efficient Gaussian Splatting},\n  author        = {Florian Hahlbohm and Jorge Condor and Linus Franke and Martin Eisemann and Marcus Magnor},\n  year          = {2026},\n  eprint        = {2609.05255},\n  archivePrefix = {arXiv},\n  primaryClass  = {cs.CV},\n  url           = {https://arxiv.org/abs/2609.05255},\n}\n```\n\n### Acknowledgements\n\nWe thank Timon Scholz for his contributions to the software infrastructure supporting this work and Jannis Möller for identifying and helping us fix a bug in tiny-cuda-nn. This work was partially funded by the DFG projects “Real-Action VR” (ID 523421583) and “Increasing Realism of Omnidirectional Videos in Virtual Reality” (ID 491805996).\n\n                The website template was adapted from [Zip-NeRF](https://jonbarron.info/zipnerf/), who\n                borrowed from [Michaël Gharbi](http://mgharbi.com/) and [Ref-NeRF](https://dorverbin.github.io/refnerf).", "url": "https://wpnews.pro/news/compact-neural-appearance-models-for-efficient-gaussian-splatting", "canonical_source": "https://fhahlbohm.github.io/efficient-gaussian-appearance/", "published_at": "2026-09-14 16:33:45+00:00", "updated_at": "2026-09-14 16:56:47.044129+00:00", "lang": "en", "topics": ["computer-vision", "neural-networks", "ai-research", "machine-learning"], "entities": ["Florian Hahlbohm", "Jorge Condor", "Linus Franke", "Martin Eisemann", "Marcus Magnor", "3D Gaussian Splatting", "arXiv", "CUDA"], "alternates": {"html": "https://wpnews.pro/news/compact-neural-appearance-models-for-efficient-gaussian-splatting", "markdown": "https://wpnews.pro/news/compact-neural-appearance-models-for-efficient-gaussian-splatting.md", "text": "https://wpnews.pro/news/compact-neural-appearance-models-for-efficient-gaussian-splatting.txt", "jsonld": "https://wpnews.pro/news/compact-neural-appearance-models-for-efficient-gaussian-splatting.jsonld"}}