Research map / Learning theory and algorithms / Optimal transport
SFBD-OMNI: Bridge models for lossy measurement restoration with limited clean samples
Haoye Lu, Yaoliang Yu, Darren Lo
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 50 clean samples as prior to SFBD-OMNI under additive Gaussian noise (ablation) | SFBD-OMNI · CIFAR-10 additive Gaussian | FID: 80.37 -> 10.81 | Improved |
| keep clean samples active as soft prior when corruption is non-identifiable (ablation) | Online SFBD-OMNI · CelebA grayscale and Gaussian blur | FID | Improved |
| use clean-sample weight in identifiable corruption (Gaussian noise) (ablation) | Online SFBD-OMNI · CIFAR-10 additive Gaussian | FID | Hurt |
| overly large clean-sample weight in non-identifiable corruption (ablation) | Online SFBD-OMNI · non-identifiable corruptions | FID | Hurt |
| increase number of clean samples (ablation) | Online SFBD-OMNI · CelebA grayscale | FID | Improved |
| larger update ratio gamma (0.5) (ablation) | Online SFBD-OMNI · CIFAR-10 additive Gaussian (sigma=0.5) | FID | Mixed |
| SFBD-OMNI instead of EMDiffusion | EMDiffusion · CIFAR-10 pixel masking | FID: 21.08 -> 21.31 | No clear effect |
| SFBD-OMNI instead of SFBD | SFBD · CIFAR-10 additive Gaussian | FID: 13.53 -> 10.81 | Improved |
| online variant instead of classical SFBD-OMNI | SFBD-OMNI · identifiable corruptions (CIFAR-10 masking, Gaussian) | FID: 21.31 -> 22.43; 10.81 -> 11.06 | Hurt |
| online variant instead of classical SFBD-OMNI | SFBD-OMNI · non-identifiable corruptions (CIFAR-10 grayscale, CelebA blur, grayscale) | FID: 32.61 -> 31.32; 11.60 -> 10.28; 11.85 -> 11.21 | Improved |
Closest papers by meaning
- Adapt and Diffuse: Sample-adaptive Reconstruction via Latent Diffusion Models (ICML 2024 · similarity 0.86)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.86)
- Stochastic Forward–Backward Deconvolution: Training Diffusion Models with Finite Noisy Datasets (ICML 2025 · similarity 0.85)
- Zeroth-Order Sampling Methods for Non-Log-Concave Distributions: Alleviating Metastability by Denoising Diffusion (NeurIPS 2024 · similarity 0.85)
- The Price of Sparsity: Sufficient Conditions for Sparse Recovery using Sparse and Sparsified Measurements (NeurIPS 2025 · similarity 0.84)
- How many measurements are enough? Bayesian recovery in inverse problems with general distributions (NeurIPS 2025 · similarity 0.84)
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.
