Research map / 3D vision / 3D generation
Dehallu3D: Hallucination-Mitigated 3D Generation from a Single Image via Cyclic View Consistency Refinement
Xiwen Wang, Shichao Zhang, Ruowei Wang, Mao Li, Chenyu Zhou, Ji-Zhe Zhou, Qijun Zhao, Hailun 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 cyclic depth consistency loss L_DC (without L_SE or L_DS) (ablation) | Dehallu3D · GSO | PSNR: 16.8972 -> 20.9276 | Improved |
| add depth smoothness loss L_DS to L_DC (ablation) | Dehallu3D without L_SE · GSO | PSNR: 20.9276 -> 21.1973 | Improved |
| add surface exposure-weighted normal loss L_SE (ablation) | Dehallu3D with L_DC and L_DS · GSO | PSNR: 21.1973 -> 21.8407 | Improved |
| add L_SE alone (ablation) | baseline reconstruction · GSO | PSNR: 16.8972 -> 17.6534 | Improved |
| add L_DS alone (ablation) | baseline reconstruction · GSO | PSNR: 16.8972 -> 18.3124 | Improved |
| reduce view angular interval from 5 to 3 degrees (ablation) | CVCR · GSO | Chamfer Distance / F-Score: F-Score 0.4212 -> 0.4203, time 163.3 -> 208.1 s | No clear effect |
| increase view angular interval from 5 to 10 degrees (ablation) | CVCR · GSO | PSNR: 21.8407 -> 21.3687 | Hurt |
| increase view angular interval from 5 to 15 degrees (ablation) | CVCR · GSO | PSNR: 21.8407 -> 20.8415 | Hurt |
| plug CVCR module into Unique3D | Unique3D · qualitative | mesh outliers | Improved |
| Dehallu3D versus SF3D, the second-best method | single-image-to-3D · GSO | F-Score: 0.3765 -> 0.4212 | 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.
- SF3D — SF3D: Stable Fast 3D Mesh Reconstruction with UV-unwrapping and Illumination Disentanglement (CVPR 2025) · better on 4 of 4 shared results
- Unique3D — Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image (NeurIPS 2024) · better on 2 of 2 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
- Hallo3D: Multi-Modal Hallucination Detection and Mitigation for Consistent 3D Content Generation (NeurIPS 2024 · similarity 0.91)
- Cycle3D: High-quality and Consistent Image-to-3D Generation via Generation-Reconstruction Cycle (AAAI 2025 · similarity 0.87)
- DMV3D: Denoising Multi-view Diffusion Using 3D Large Reconstruction Model (ICLR 2024 · similarity 0.87)
- DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior (ICLR 2024 · similarity 0.86)
- Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image (NeurIPS 2024 · similarity 0.86)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · similarity 0.86)
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.
