FineSplat: Fine Patches, Local Groups, Detailed Splats

Key points

• Yet, fine-grained geometry and appearance remain difficult to recover, motivating us to revisit how feed-forward models represent, aggregate, and refine detailed visual information.1
• We introduce FineSplat, a feed-forward 3DGS framework built around fine-grained image patches.1
• Since global multi-view attention becomes prohibitively expensive at such high token resolutions, we develop a dual-grouping strategy that combines two complementary forms of locality: view-wise grouping enables efficient interaction among neighboring views, while 3D spatial grouping reconnects geometrically related tokens that may never have interacted under view-wise grouping by lifting them into the 3D scene space.1
• We further leverage reconstruction error as direct feedback, providing an explicit rendering-based signal to recover fine-grained structures and appearance, thereby overcoming the limitations of one-pass feed-forward prediction.1

Sources (2)

• [1]FineSplat: Fine Patches, Local Groups, Detailed Splats
  Hugging Face Daily Papers · Oct 18, 12:00 AM

  Feed-forward 3D Gaussian Splatting (3DGS) promises instant scene reconstruction with high-quality novel-view synthesis.

  Yet, fine-grained geometry and appearance remain difficult to recover, motivating us to revisit how feed-forward models represent, aggregate, and refine detailed visual information.

• [2]FineSplat: Fine Patches, Local Groups, Detailed Splats
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 04:44 AM · same content

Extractive summary: sentences quoted from the sources.