Research map / Video / Video generation
Ouroboros: Single-step Diffusion Models for Cycle-consistent Forward and Inverse Rendering
Shanlin Sun, Yifan Wang, Hanwen Zhang, Yifeng Xiong, Qin Ren, Ruogu Fang, Xiaohui Xie, Chenyu You
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 |
|---|---|---|---|
| single-step end-to-end finetuned diffusion instead of multi-step RGB↔X | RGB↔X · inference speed | speedup: 50× | Improved |
| add cycle training and outdoor data finetuning to single-step model | Ouroboros · MatrixCity albedo | PSNR: 12.61 -> 25.38 | Improved |
| single-step finetuned model vs RGB↔X | Ouroboros · Hypersim normal | mean angular error: 17.21 -> 11.98 | Improved |
| single-step finetuned model vs RGB↔X | Ouroboros · MatrixCity forward rendering | PSNR: 9.24 -> 21.57 | Improved |
| single-step finetuned model vs RGB↔X | Ouroboros · Hypersim forward rendering | LPIPS: 0.20 -> 0.25 | Hurt |
| single-step model vs RGB↔X | Ouroboros · Hypersim irradiance | PSNR: 11.64 -> 12.07 | Improved |
| single-step model vs RGB↔X | Ouroboros · Hypersim albedo | LPIPS: 0.20 -> 0.23 | Hurt |
| single-step model vs E2E | Ouroboros · MatrixCity normal | mean angular error: 13.91 -> 18.12 | Hurt |
| add cycle training (ablation) | Ouroboros · outdoor RGB-to-RGB reconstruction (qualitative) | visual fidelity | Improved |
| add wild data (MSCOCO, Flickr30k) to cycle training (ablation) | Ouroboros · irradiance and metallicity (qualitative) | visual quality | Improved |
| add e2e loss in cycle training (ablation) | Ouroboros · metallicity and irradiance (qualitative) | visual quality | Improved |
Closest papers by meaning
- Diffusion Renderer: Neural Inverse and Forward Rendering with Video Diffusion Models (CVPR 2025 · similarity 0.91)
- Channel-wise Noise Scheduled Diffusion for Inverse Rendering in Indoor Scenes (CVPR 2025 · similarity 0.90)
- Uni-Renderer: Unifying Rendering and Inverse Rendering Via Dual Stream Diffusion (CVPR 2025 · similarity 0.90)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · similarity 0.88)
- Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion (AAAI 2025 · similarity 0.88)
- Inverse Rendering using Multi-Bounce Path Tracing and Reservoir Sampling (ICLR 2025 · similarity 0.88)
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.
