Research map / 3D vision / 3D generation
AlignCVC: Aligning Cross-View Consistency for Single-Image-to-3D Generation
Xinyue Liang, Zhiyuan Ma, Lingchen Sun, Yanjun Guo, Lei Zhang
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 score-distillation soft alignment for the MVG model (ablation) | AlignCVC (Wonder3D+LGM) · GSO | PSNR: 18.28 -> 21.98 | Improved |
| add adversarial hard alignment for the reconstruction model (ablation) | AlignCVC (Wonder3D+LGM) · GSO | PSNR: 20.84 -> 21.98 | Improved |
| add regression loss LRecon alongside distribution alignment (ablation) | AlignCVC (Wonder3D+LGM) · GSO | PSNR: 21.04 -> 21.98 | Improved |
| joint training of MVG and reconstruction instead of fixing either (ablation) | AlignCVC (Wonder3D+LGM) · GSO | PSNR: Fixed MVG 17.40, Fixed Recon. 19.05 -> 21.98 | Improved |
| apply 3D-aware sampling to pre-trained models without alignment training (ablation) | Wonder3D+LGM · GSO | FID: 189.97 -> 249.05 | Hurt |
| fine-tune only the reconstruction model on inconsistent generated views (ablation) | Wonder3D+LGM two-stage · GSO | PSNR / SSIM / FID: PSNR 16.89 -> 18.24; SSIM 0.72 -> 0.66; FID 189.97 -> 194.44 | Mixed |
| use adversarial hard alignment for the MVG model instead of score distillation (ablation) | AlignCVC · GSO | PSNR: 21.98 -> 17.31 | Hurt |
| use score distillation (soft alignment) for the reconstruction model instead of GAN (ablation) | AlignCVC · GSO | PSNR: 21.98 -> 16.69 | Hurt |
| replace distillation loss with diffusion loss for MVG (ablation) | AlignCVC · GSO | PSNR: 21.98 -> 16.39 | Hurt |
| apply AlignCVC alignment training to pre-trained MVG+reconstruction pairs | Wonder3D+LaRa · GSO | FID: 220.770 -> 158.699 | Improved |
| apply AlignCVC with 4-step inference vs prior 3D-aware sampling | Wonder3D+LGM · GSO | inference time (s): 41.52 (Gen-3Diffusion) -> 8.87 | Improved |
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.
- Ouroboros3D — Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025) · better on 3 of 3 shared results
- SyncDreamer — SyncDreamer: Generating Multiview-consistent Images from a Single-view Image (ICLR 2024) · better on 1 of 1 shared result
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
- SynCamMaster: Synchronizing Multi-Camera Video Generation from Diverse Viewpoints (ICLR 2025 · similarity 0.88)
- MET3R: Measuring Multi-View Consistency in Generated Images (CVPR 2025 · similarity 0.88)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · similarity 0.88)
- Flex3D: Feed-Forward 3D Generation with Flexible Reconstruction Model and Input View Curation (ICML 2025 · similarity 0.88)
- GenFusion: Closing the Loop between Reconstruction and Generation via Videos (CVPR 2025 · similarity 0.87)
- HarmonyView: Harmonizing Consistency and Diversity in One-Image-to-3D (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.
