Research map / 3D vision / Human pose estimation
Reconstruction by Generation: 3D Multi-Object Scene Reconstruction from Sparse Observations
Andrii Zadaianchuk, Leonardo Barcellona, Lennard Schuenemann, Christian Gumbsch, Zehao Wang, Muhammad Zubair Irshad, Fabien Despinoy, Rahaf Aljundi, Efstratios Gavves, Sergey Zakharov
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 stereo-noise depth augmentation during training (ablation) | RecGen · object-centric (HB, LM-O, ReOcS) | ADD-SB: 0.078 -> 0.062 | Improved |
| add stereo-noise augmentation (ablation) | RecGen · ArtVIP parts (synthetic) | ADD-SB: 0.039 vs 0.043 | No clear effect |
| add pose z-score normalization (ablation) | RecGen · object-centric | ADD-SB: 0.074 -> 0.062 | Improved |
| add pose normalization (ablation) | RecGen · ArtVIP parts | ADD-SB: 0.056 -> 0.043 | Improved |
| add part-level training data (ablation) | RecGen · ArtVIP parts | CD_norm: 0.073 -> 0.033 | Improved |
| add part-level training data (ablation) | RecGen · object-centric | ADD-SB: 0.060 vs 0.062 | No clear effect |
| initialize from pretrained TRELLIS weights (ablation) | RecGen · object-centric | ADD-SB: 0.067 -> 0.062 | Improved |
| use 2-view instead of 1-view input | RecGen · LM-O | CD_norm: 0.050 -> 0.056 | Hurt |
| use 2-view instead of 1-view input | RecGen · HB | CD_norm: 0.032 -> 0.029 | Improved |
| joint shape-pose generation with pointmap conditioning vs SAM3D | RecGen vs SAM3D · ArtVIP parts | ADD-SB@0.05: 45.8% -> 84.0% | Improved |
| joint shape-pose generation vs SAM3D | RecGen vs SAM3D · HB | ADD-SB@0.05: 54.6% -> 73.8% | Improved |
| pick best of 5 samples by pointmap alignment | RecGen · HB | CD_n: 0.031 -> 0.029 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAM3D — SAM 3D: 3Dfy Anything in Images (CVPR 2026) · better on 9 of 9 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
- PartGen: Part-level 3D Generation and Reconstruction with Multi-view Diffusion Models (CVPR 2025 · similarity 0.88)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.88)
- AutoPartGen: Autoregressive 3D Part Generation and Discovery (NeurIPS 2025 · similarity 0.88)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.87)
- Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis (NeurIPS 2024 · similarity 0.87)
- G3DR: Generative 3D Reconstruction in ImageNet (CVPR 2024 · similarity 0.87)
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.
