Research map / Image generation / Fast diffusion sampling
Trajectory Stitching for Solving Inverse Problems with Flow-Based Models
Alexander Denker, Zeljko Kereta, Carola-Bibiane Schönlieb, Moshe Eliasof
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 |
|---|---|---|---|
| replace single-shooting D-Flow with multiple-shooting MS-Flow | D-Flow · CelebA Gaussian deblurring | PSNR (dB): 36.07 -> 39.04 | Improved |
| replace D-Flow with MS-Flow | D-Flow · CelebA 2x super-resolution | PSNR (dB): 35.49 -> 36.41 | Improved |
| replace D-Flow with MS-Flow | D-Flow · CelebA 4x super-resolution | PSNR (dB): 29.93 -> 28.52 | Hurt |
| replace D-Flow with MS-Flow | D-Flow · CelebA box-inpainting | PSNR (dB): 34.20 -> 34.77 | Mixed |
| replace D-Flow with MS-Flow | D-Flow · CelebA Gaussian deblurring | SSIM: 0.957 -> 0.968 | Mixed |
| replace D-Flow with MS-Flow | D-Flow · CelebA image recovery | time per image (s): 26.31 -> 14.11 | Improved |
| use MS-Flow with JFB (L=1) instead of D-Flow at n_t=12 | D-Flow · OrganCMNIST CT | computation time (s): 175.57 -> 4.44 | Improved |
| add Jacobian-free approximation to MS-Flow (L=1, n_t=12) (ablation) | MS-Flow w/o JFB · OrganCMNIST CT | computation time (s): 41.05 -> 4.44 | Improved |
| increase inner iterations L from 1 to 10 (n_t=12) (ablation) | MS-Flow w/ JFB · OrganCMNIST CT | computation time (s): 4.44 -> 40.10 | Hurt |
| use exact gradients instead of Jacobian-free approximation (ablation) | MS-Flow coordinate descent · OrganCMNIST CT | trajectory loss convergence | No clear effect |
| replace FlowDPS with MS-Flow | FlowDPS · FFHQ deblurring noise 0.01 | PSNR (dB): 27.78 -> 30.65 | Improved |
| replace FLAIR with MS-Flow | FLAIR · FFHQ deblurring noise 0.01 | LPIPS: 0.199 -> 0.268 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- D-Flow — DFlow: A Generative Model Combining Denoising AutoEncoder and Normalizing Flow for High Fidelity Waveform Generation (ICML 2024) · better on 7 of 8 shared results
- PnP-Flow — PnP-Flow: Plug-and-Play Image Restoration with Flow Matching (ICLR 2025) · better on 3 of 5 shared results
- FlowDPS — FlowDPS : Flow-Driven Posterior Sampling for Inverse Problems (ICCV 2025) · better on 3 of 3 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
- Flow Priors for Linear Inverse Problems via Iterative Corrupted Trajectory Matching (NeurIPS 2024 · similarity 0.92)
- Solving Inverse Problems with FLAIR (NeurIPS 2025 · similarity 0.92)
- Latent Refinement via Flow Matching for Training-free Linear Inverse Problem Solving (NeurIPS 2025 · similarity 0.91)
- Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations (ICLR 2025 · similarity 0.88)
- Text-to-Image Rectified Flow as Plug-and-Play Priors (ICLR 2025 · similarity 0.88)
- Fast samplers for Inverse Problems in Iterative Refinement models (NeurIPS 2024 · 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.
