PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction
kwangmoo_yi · x · 2026-10-27
A paper from Imperial College London and University of Tübingen, PixGR (code released), integrates pixel-space generative decoding into fast feed-forward 3D reconstruction.
Approach: Context observations are encoded into a scene latent, which conditions generative denoising at noisy target views, jointly producing a shared 3D Gaussian representation — denoising target views and generating 3D Gaussians simultaneously.
Advantages:
- vs. regression-based decoding: deterministic regressors struggle with large viewpoint changes and occlusion; introducing noisy target views improves interpolation with little context overlap and strengthens extrapolation in occluded regions.
- vs. two-stage latent-generation models: staged pipelines propagate errors and VAE latents lose fine detail; PixGR learns generation and reconstruction end-to-end in pixel space.
Denoising more target views improves reconstruction in weakly observed regions, with better geometry and appearance as target count grows.
More from Multimodal
- Runway App Studio Demos: Character Swap and Production-Ready Character Sheets — runwayml · 2026-10-27
- Higgsfield Launches Playable: One-Prompt Playable Ads — CurieuxExplorer · 2026-10-27
- Guy turns his girlfriend's selfie into a playable AI game ad about popping pimples — CurieuxExplorer · 2026-10-27
- Fibo 2 edit called unstable: even a color change alters the edited object — jd3k · 2026-10-27
- AIArtsy AI generates 3D cartoon-style sword fighting animation clip — Jealous-Associate-41 · 2026-10-27
- Seeking AI workflows for high-quality anime comics and colored manga pages — Worldly_Courage395 · 2026-10-27