Research map / Image generation / Fast diffusion sampling
Resolving Endpoint Underfitting in Diffusion Bridges via Noise Alignment
Yurong Gao, Zicheng Zhang, Congying Han, Tiande Guo, Xinmin Qiu
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 Mean Network to NADB (ablation) | NADB · JPEG QF=5 NFE=100 | FID: 5.2 -> 4.3 | Improved |
| add Mean Network to NADB (ablation) | NADB · JPEG QF=5 NFE=10 | FID: 7.0 -> 6.9 | No clear effect |
| add Mean Network to I2SB without magnitude alignment (ablation) | I2SB · JPEG QF=5 NFE=10 | FID: 8.0 -> 8.8 | Hurt |
| add Mean Network to I2SB without magnitude alignment (ablation) | I2SB · JPEG QF=5 NFE=100 | FID: 4.7 -> 4.4 | Improved |
| add magnitude-aligned interpolant on top of Mean Network (ablation) | I2SB w. Mean · JPEG QF=5 NFE=10 | FID: 8.8 -> 6.9 | Improved |
| alpha 0.3 to 0.4 (ablation) | NADB · JPEG QF=5 NFE=10 | FID: 8.2 -> 6.9 | Improved |
| alpha 0.4 to 0.5 (ablation) | NADB · JPEG QF=5 NFE=10 | FID: 6.9 -> 7.1 | Hurt |
| replace I2SB with NADB | I2SB · Deblurring Uniform NFE=10 | FID: 10.3 -> 4.8 | Improved |
| replace I2SB with NADB | I2SB · Deblurring Gaussian NFE=10 | PSNR: 25.42 -> 30.03 | Improved |
| replace I2SB with NADB | I2SB · JPEG QF=5 NFE=10 | PSNR: 24.50 -> 24.45 | No clear effect |
| replace I2SB with NADB | I2SB · 4x SR Bicubic NFE=100 | FID: 4.1 -> 4.1 | No clear effect |
| replace PMRF with NADB at higher NFE | PMRF · JPEG QF=5 NFE=100 | FID: 7.8 -> 4.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- I2SB — I$^2$SB: Image-to-Image Schrödinger Bridge (ICML 2023) · better on 13 of 14 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
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.91)
- Diffusion Bridge Implicit Models (ICLR 2025 · similarity 0.89)
- Underdamped Diffusion Bridges with Applications to Sampling (ICLR 2025 · similarity 0.89)
- Exploring the Design Space of Diffusion Bridge Models (NeurIPS 2025 · similarity 0.89)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.88)
- UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control (ICML 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.
