Research map / Image generation / Fast diffusion sampling
Projected Coupled Diffusion for Test-Time Constrained Joint Generation
Hao Luan, Yi Xian Goh, See-Kiong Ng, Chun Kai Ling
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 projection to enforce velocity limits (ablation) | Diffuser · Multi-robot Empty (4 robots, max vel 0.703) | min constraint satisfaction (%): 62.3 -> 100 | Improved |
| add coupling cost on top of projection (ablation) | Diffuser + proj. · Multi-robot Empty (4 robots, max vel 0.703) | success rate (%): 65.0 -> 100 | Improved |
| add projection to coupled diffusion (CD-SHD to PCD-SHD) (ablation) | coupled diffusion · Multi-robot Empty (4 robots, max vel 0.703) | min constraint satisfaction (%): 35.3 -> 100 | Improved |
| add projection to coupled diffusion (log-barrier) (ablation) | coupled diffusion · Multi-robot Highways (4 robots, max vel 0.878) | min constraint satisfaction (%): 0.00 -> 100 | Improved |
| use log-barrier coupling instead of squared hinge cost (ablation) | PCD · Multi-robot Empty (4 robots, max vel 0.703) | data adherence minDA: 0.960 -> 0.489 | Hurt |
| add coupling cost to projected Diffusion Policy (ablation) | DP + proj. · PushT | DTW between pair: 3.0 -> 5.1 (PCD-LB) | Improved |
| add projection with coupling (velocity limit) (ablation) | DP with coupling · PushT | task completion TC: 0.91 -> 0.78 (CD-LB to PCD-LB) | Hurt |
| use posterior sampling variant of coupling cost with projection (ablation) | PCD-LB · PushT | task completion TC: 0.78 -> 0.89 | Improved |
| add coupling cost to projection-only SD (ablation) | SD+P · paired faces | age-contrast XOR (%): 44 -> 96 | Improved |
| add projection to Stable Diffusion (ablation) | SD · paired faces | gender satisfaction M/F: 71/14 -> 100/100 (SD+P) | Improved |
| add projection to Stable Diffusion (ablation) | SD · paired faces | intra-sample diversity IS-LPIPS: .75 -> .06 (SD+P) | Hurt |
| increase coupling strength (ablation) | PCD · multi-robot / general | data fidelity | Hurt |
Closest papers by meaning
- Constrained Diffusion with Trust Sampling (NeurIPS 2024 · similarity 0.89)
- Constrained Generative Modeling with Manually Bridged Diffusion Models (AAAI 2025 · similarity 0.89)
- Constrained Synthesis with Projected Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Model-based Diffusion for Trajectory Optimization (NeurIPS 2024 · similarity 0.87)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.87)
- Generative Trajectory Stitching through Diffusion Composition (NeurIPS 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.
