Research map / 3D vision / 3D generation
CyC3D: Fine-grained Controllable 3D Generation via Cycle Consistency Regularization
Hongbin Xu, Chaohui Yu, Feng Xiao, Jiazheng Xing, Hai Ci, Weitao Chen, Fan Wang, Ming Li
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 2D condition cycle consistency loss L_cond (ablation) | CyC3D-D · GSO edge | PSNR: 12.87 -> 15.09 | Improved |
| Add 2D condition cycle consistency loss L_cond (ablation) | CyC3D-D · GSO sketch | PSNR: 17.52 -> 18.38 | Improved |
| Add 2D cycle consistency loss L_cond (ablation) | CyC3D-D · GSO depth | R-MSE: 0.0650 -> 0.0321 | Improved |
| Add 3D depth/normal condition cycle loss L_cond-d on top of L_cond (ablation) | CyC3D-D · GSO depth | R-MSE: 0.0321 -> 0.0060 | Improved |
| Add view cycle consistency (render + CLIP semantic loss) (ablation) | CyC3D · controllable 3D generation (qualitative) | 3D shape quality | Improved |
| Train with cycle consistency (CyC3D-D vs ControLRM-D backbone) | CyC3D-D · GSO depth | R-MSE: 0.0650 -> 0.0060 | Improved |
| Train with cycle consistency (CyC3D-D vs ControLRM-D) | CyC3D-D · GSO edge | PSNR: 12.47 -> 15.09 | Improved |
| Train with cycle consistency (CyC3D-D vs ControLRM-D) | CyC3D-D · GOBJ to GSO generation quality | FID: 171.13 -> 168.71 | No clear effect |
| Train with cycle consistency (CyC3D-T vs ControLRM-T) | CyC3D-T · GOBJ to GSO generation quality | FID: 260.75 -> 255.30 | No clear effect |
| Use transformer backbone (CyC3D-T) vs diffusion backbone (CyC3D-D) | CyC3D · GOBJ to GSO generation quality | FID: 168.71 (D) vs 255.30 (T) | Hurt |
Closest papers by meaning
- Sculpt3D: Multi-View Consistent Text-to-3D Generation with Sparse 3D Prior (CVPR 2024 · similarity 0.88)
- Interactive3D: Create What You Want by Interactive 3D Generation (CVPR 2024 · similarity 0.88)
- Eval3D: Interpretable and Fine-grained Evaluation for 3D Generation (CVPR 2025 · similarity 0.87)
- Consistent3D: Towards Consistent High-Fidelity Text-to-3D Generation with Deterministic Sampling Prior (CVPR 2024 · similarity 0.87)
- Cycle3D: High-quality and Consistent Image-to-3D Generation via Generation-Reconstruction Cycle (AAAI 2025 · similarity 0.87)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · 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.
