DensiTok: flow-matching token densification lets frozen feed-forward 3DGS see unseen views
zhenjun_zhao · x · 2026-10-08
DensiTok is a plug-in module for pretrained feed-forward 3D Gaussian Splatting models. It compresses internal geometry tokens into a compact latent space, completes latents for unobserved viewpoints in a single flow-matching step conditioned on camera geometry, and decodes them back for the original reconstruction heads—with backbone and heads frozen. Unlike view-synthesis re-encoding pipelines, it needs no image generation and stays 3D-consistent, reducing holes, floaters and blur in sparse-view settings.
Related event: DensiTok: Completing Geometry Tokens for Sparse-View 3DGS(2 posts)→
More from Research
- EMNLP 2026 paper RECAP trains reasoning models to recover from unsafe trajectories — pinyuchenTW · 2026-10-08
- SaTML 2027 to host AdvML Frontiers workshop on human-centered trustworthy machine learning — pinyuchenTW · 2026-10-08
- OpenAI's new result proves 2005 edit-distance embedding optimal; researcher distills proof to 2.5 pages with AI help — thegautamkamath · 2026-10-08
- Experiments show smarter models and higher effort write better LLM-judge evals — danshipper · 2026-10-08
- Tencent's WorkForge scales verifiable training environments for long-horizon work agents — teortaxesTex · 2026-10-08
- Masked Geometric Encoder boosts 3D foundation models via frame dropping and self-distillation — zhenjun_zhao · 2026-10-08