Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting

2026-03-28 · Source: Apple Machine Learning Research · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

A new feed-forward 3D Gaussian Splatting framework, LGTM (Less Gaussians, Texture More), addresses the scalability limitations of existing methods in high-resolution novel view synthesis. Published in March 2026 for ICLR, LGTM decouples geometric complexity from rendering resolution by predicting compact Gaussian primitives combined with per-primitive textures. This innovation allows for high-fidelity 4K novel view synthesis without requiring per-scene optimization, a significant advancement over previous feed-forward techniques. The method achieves this while utilizing substantially fewer Gaussian primitives, making 4K synthesis tractable for the first time in this domain.

Key takeaway

For research scientists developing real-time 3D rendering systems, LGTM offers a pathway to overcome resolution scaling barriers. You should investigate integrating per-primitive textures with compact Gaussian representations to achieve high-fidelity 4K output without the computational overhead of traditional Gaussian Splatting, potentially enabling new applications in high-resolution virtual environments or digital twins.

Key insights

LGTM enables 4K novel view synthesis by decoupling geometry from resolution using textured Gaussian primitives.

Principles

• Decouple geometry from rendering resolution.
• Compact primitives with textures improve scalability.

Method

LGTM predicts compact Gaussian primitives and couples them with per-primitive textures to enable high-resolution synthesis without quadratic primitive growth.

In practice

• Achieve 4K novel view synthesis.
• Reduce Gaussian primitive count significantly.

Topics

• LGTM
• 3D Gaussian Splatting
• Novel View Synthesis
• 4K Resolution
• Computer Vision

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Machine Learning Engineer

Related on AIssential

• Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting — LGTM uses textured Gaussian primitives to enable 4K novel view synthesis with fewer primitives.
• Local-GS: Accelerating 3D Gaussian Splatting via Tile-Local Warp Coherence — Local-GS enhances 3DGS rendering by aligning Gaussian processing with GPU warp execution for improved efficiency.
• RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery — Native RPC integration in Gaussian Splatting significantly improves satellite imagery 3D reconstruction accuracy.
• Gaussian Light Field Splatting: A Physical Prior-Driven Vision Transformer for Unsupervised Low-Light Image Enhancement — GLFS integrates Gaussian splatting's physical illumination modeling into a Vision Transformer for unsupervised low-light image enhancement.
• WildSplatter: Feed-forward 3D Gaussian Splatting with Appearance Control from Unconstrained Images — WildSplatter enables fast, pose-free 3DGS from unconstrained images by disentangling geometry and appearance via learned embeddings.
• Feed-Forward 3D Scene Modeling: A Problem-Driven Perspective — Feed-forward 3D reconstruction offers efficient, generalizable scene modeling by directly mapping 2D inputs to 3D representations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Apple Machine Learning Research.