Research map / Image generation / Fast diffusion sampling
Structured Diffusion Bridges: Inductive Bias for Denoising Diffusion Bridges
Eitan Kosman, Gabriele Serussi, Chaim Baskin
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 endpoint cycle consistency to marginal matching (ablation) | SDB · Synthetic benchmark (rho=0) | content accuracy: 0.162 -> 0.662 | Improved |
| add trajectory-level cycle consistency on top of endpoint cycle (ablation) | SDB · Synthetic benchmark (rho=0) | content accuracy: 0.662 -> 0.868333 | Improved |
| add trajectory cycle on top of endpoint cycle (ablation) | SDB · Synthetic benchmark (rho=0) | cycle MSE: 0.831 -> 0.679605 | Improved |
| add structural constraints to paired data (semi-paired vs paired-only) (ablation) | SDB · Synthetic benchmark (rho=0.5) | content accuracy: 0.641 -> 0.955 | Improved |
| add structural constraints to fully paired data (ablation) | SDB · Synthetic benchmark (rho=1.0) | content accuracy: 0.887 -> 0.965 | Improved |
| add endpoint and trajectory cycle to marginal matching (ablation) | SDB · Zero-shot SR CelebA-HQ (rho=0) | PSNR: 17.8 -> 19.0 | Improved |
| add semi-paired structured training to paired-only (ablation) | SDB · Zero-shot SR CelebA-HQ (rho=0.5) | PSNR: 24.9 -> 25.2 | Improved |
| use SDB instead of LDDBM (ablation) | bridge model · ShapeNet multiview-to-3D (rho=0.5) | 1-NNA: 0.660 -> 0.559 | Improved |
| use SDB instead of LRGT+ at full pairing (ablation) | multiview-to-3D · ShapeNet multiview-to-3D (rho=1.0) | IoU: 0.762 -> 0.677 | Hurt |
| set WTA capacity constraint C_y=2 instead of unconstrained (ablation) | SDB · Synthetic benchmark | content accuracy: 0.921 -> 0.955 | Improved |
| increase WTA candidates K from 1 to 16 (ablation) | SDB · Synthetic benchmark | content accuracy: 0.845 -> 0.971 | Improved |
| learn bridge directly in pixel space instead of autoencoder latent space (ablation) | SDB · Super-resolution | qualitative | Hurt |
Closest papers by meaning
- Towards General Modality Translation with Contrastive and Predictive Latent Diffusion Bridge (NeurIPS 2025 · similarity 0.94)
- Diffuse Everything: Multimodal Diffusion Models on Arbitrary State Spaces (ICML 2025 · similarity 0.88)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.88)
- The Design Space of Tri-Modal Masked Diffusion Models (NeurIPS 2026 · similarity 0.88)
- Where to Look Is Not How to Fix: Pre-Denoising Diagnostics and Modality-Dependent Control in Diffusion Composition (NeurIPS 2026 · similarity 0.88)
- Flowing from Words to Pixels: A Noise-Free Framework for Cross-Modality Evolution (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.
