Current feed-forward 3D Gaussian Splatting strategies predict pixel-aligned primitives, resulting in a quadratic progress in primitive rely as decision will increase. This basically limits their scalability, making high-resolution synthesis corresponding to 4K intractable. We introduce LGTM (Much less Gaussians, Texture Extra), a feed-forward framework that overcomes this decision scaling barrier. By predicting compact Gaussian primitives coupled with per-primitive textures, LGTM decouples geometric complexity from rendering decision. This method allows high-fidelity 4K novel view synthesis with out per-scene optimization, a functionality beforehand out of attain for feed-forward strategies, all whereas utilizing considerably fewer Gaussian primitives.
- † The College of Hong Kong
- ** Work achieved whereas at Apple
