Research map / Image generation / Editing and style transfer
BBDM: Image-to-Image Translation With Brownian Bridge Diffusion Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Model translation as a direct Brownian Bridge between domains rather than conditional generation | BBDM · CelebAMask-HQ, edges2shoes, edges2handbags, faces2comics | FID: CelebAMask-HQ: 22.816 -> 21.350; edges2shoes: 13.020 -> 10.924; edges2handbags: 24.251 -> 17.257; faces2comics: 24.280 - | Helped |
| Use BBDM instead of conditional diffusion models | BBDM · edges2shoes | LPIPS: 0.173 -> 0.183 | Mixed |
| Use VQGAN-f8 latent space instead of VQGAN-f4 (ablation) | BBDM · CelebAMask-HQ | FID: 21.350 -> 21.966 | Hurt |
| Use VQGAN-f16 latent space instead of VQGAN-f4 (ablation) | BBDM · CelebAMask-HQ | FID: 21.350 -> 22.061 | Hurt |
| Use different VQGAN latent downsampling factors (ablation) | BBDM · CelebAMask-HQ | FID: 21.350; 21.966; 22.061 | No clear effect |
| Increase reverse sampling steps from 20 to 200 (ablation) | BBDM · CelebAMask-HQ | FID: 33.409 -> 21.350 | Helped |
| Increase reverse sampling steps from 200 to 1000 (ablation) | BBDM · CelebAMask-HQ | FID: 21.350 -> 21.348 | Helped |
| Increase reverse sampling steps from 200 to 1000 (ablation) | BBDM · CelebAMask-HQ | LPIPS: 0.370 -> 0.375 | No clear effect |
| Increase maximum-variance scale s from 1 to 2 or 4 (ablation) | BBDM · CelebAMask-HQ | Diversity: 29.859 -> 37.063; 29.859 -> 39.573 | Helped |
| Increase maximum-variance scale s from 1 to 2 or 4 (ablation) | BBDM · CelebAMask-HQ | FID: 21.350 -> 23.278; 21.350 -> 24.490 | Hurt |
| Set maximum-variance scale s to 0.5 instead of 1 (ablation) | BBDM · CelebAMask-HQ | FID: 21.350 -> 22.627 | Hurt |
Closest papers by meaning
- Inverse Bridge Matching Distillation (ICML 2025 · similarity 0.89)
- Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators (CVPR 2025 · similarity 0.89)
- Exploring the Design Space of Diffusion Bridge Models (NeurIPS 2025 · similarity 0.87)
- Diffusion Bridge Implicit Models (ICLR 2025 · similarity 0.87)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.87)
- Neural Diffusion Models (ICML 2024 · 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.