Research map / Image generation / Editing and style transfer
LBM: Latent Bridge Matching for Fast Image-to-Image Translation
Clément Chadebec, Onur Tasar, Sanjeev Sreetharan, Benjamin Aubin
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 |
|---|---|---|---|
| raise pixel loss weight from 1 to 10 (ablation) | LBM · RORD object removal | FID: 30.21 -> 26.70 | Improved |
| raise pixel loss weight from 10 to 20 (ablation) | LBM · RORD object removal | FID: 26.70 -> 27.09 | No clear effect |
| raise pixel loss weight from 1 to 10 (ablation) | LBM · RORD object removal | PSNR: 22.55 -> 22.39 | Hurt |
| add noise sigma (stochastic bridge) instead of flow matching, small sigma (ablation) | LBM · RORD object removal | FID | Improved |
| use too large sigma (ablation) | LBM · RORD object removal | FID | Hurt |
| use 4 discrete timesteps instead of uniform timestep distribution (ablation) | LBM · RORD object removal | FID | Improved |
| increase inference steps beyond 4 with discrete timestep training (ablation) | LBM · RORD object removal | FID | Hurt |
| pretrained initialization instead of training from scratch (ablation) | LBM · RORD object removal (coarse masks) | FID: 27.9 -> 26.9 | Improved |
| add synthetic rendered data to relighting training (ablation) | LBM · Image relighting | FID | Improved |
| use too much synthetic data (ablation) | LBM · Image relighting | FID | Hurt |
| operate in latent space instead of pixel space (1024 resolution) (ablation) | bridge model · Inference latency | latency: 20.11 -> 0.27 | Improved |
| operate in latent space instead of pixel space (training, 256) (ablation) | bridge model · Training peak memory | peak memory (GB): 43.19 -> 24.85 | 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.
- Harmonizer — Harmonizer: Learning to Perform White-Box Image and Video Harmonization (ECCV 2022) · better on 1 of 1 shared result
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
- BBDM: Image-to-Image Translation With Brownian Bridge Diffusion Models (CVPR 2023 · similarity 0.88)
- BiBBDM: Bidirectional Image Translation With Brownian Bridge Diffusion Models (TPAMI 2025 · similarity 0.87)
- A Diffusion Model Translator for Efficient Image-to-Image Translation (TPAMI 2024 · similarity 0.87)
- Inverse Bridge Matching Distillation (ICML 2025 · similarity 0.86)
- DBMSolver: A Training-free Diffusion Bridge Sampler for High-Quality Image-to-Image Translation (CVPR 2026 · similarity 0.86)
- Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators (CVPR 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.
