Research map / 3D vision / 3D generation
Gen3R: 3D Scene Generation Meets Feed-Forward Reconstruction
Jiaxin Huang, Yuanbo Yang, Bangbang Yang, Lin Ma, Yuewen Ma, Yiyi Liao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add joint RGB and geometry generation instead of a 2-stage RGB-then-VGGT pipeline (ablation) | Gen3R · RealEstate10K (1-view) | PSNR: 17.38 -> 20.51 | Improved |
| add joint RGB and geometry generation instead of a 2-stage pipeline (ablation) | Gen3R · Co3Dv2 (1-view geometry) | Chamfer distance: 1.6223 -> 1.1047 | Improved |
| add joint generation instead of 2-stage (ablation) | Gen3R · RealEstate10K camera control (1-view) | AUC@30: 0.6832 -> 0.7443 | Improved |
| add KL distribution alignment loss between geometry and appearance latents (ablation) | Gen3R · RealEstate10K (1-view) | PSNR: 16.31 -> 20.51 | Improved |
| add KL distribution alignment loss (ablation) | Gen3R · RealEstate10K camera control (1-view) | AUC@30: 0.4100 -> 0.7443 | Improved |
| add KL distribution alignment loss (ablation) | Gen3R · Co3Dv2 (2-view geometry) | Chamfer distance: 1.7144 -> 0.9767 | Improved |
| use adapter on VGGT tokens instead of WVD-style pretrained RGB VAE for point clouds (ablation) | geometry VAE · Co3Dv2 (feed-forward reconstruction) | Chamfer distance: 1.2080 -> 0.9986 | Improved |
| add diffusion generative model on top of the VAE for reconstruction (ablation) | Gen3R · TartanAir (feed-forward reconstruction) | Chamfer distance: 1.6518 -> 1.5101 | Improved |
| add generative prior to VGGT reconstruction | Gen3R · WildRGB-D (feed-forward reconstruction) | Chamfer distance: 0.1165 -> 0.1260 | Hurt |
| add generation of novel-view geometry to VGGT with 1 or 2 views | VGGT · Co3Dv2 (1-view geometry) | Chamfer distance: 2.3561 -> 1.1047 | Improved |
| use Gen3R instead of LVSM in 2-view appearance generation | RealEstate10K (2-view) · RealEstate10K (2-view) | PSNR: 29.58 -> 27.05 | Hurt |
Later papers that compared with this method
1 later accepted paper reports Gen3R as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- OneWorld: Taming Scene Generation with 3D Unified Representation Autoencoder (ECCV 2026 · better on 4 of 4 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 5 of 5 shared results
- Gen3C — GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (CVPR 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GenFusion: Closing the Loop between Reconstruction and Generation via Videos (CVPR 2025 · similarity 0.92)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.91)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.91)
- G3DR: Generative 3D Reconstruction in ImageNet (CVPR 2024 · similarity 0.91)
- GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (CVPR 2025 · similarity 0.90)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
