Research map / Image generation / Fast diffusion sampling
Diffusion Prior-Based Amortized Variational Inference for Noisy Inverse Problems
Sojin Lee, Dogyun Park, Inho Kong, Hyunwoo J. Kim
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 integral KL divergence (IKL) loss (ablation) | DAVI · FFHQ Gaussian deblur | FID: 50.69 -> 37.61 | Improved |
| add Perturbed Posterior Bridge with no perturbation (h=0) (ablation) | DAVI (IKL only) · FFHQ Gaussian deblur | FID: 37.61 -> 49.37 | Hurt |
| add perturbation h=0.01 to the bridge (ablation) | DAVI · FFHQ Gaussian deblur | FID: 49.37 -> 47.41 | Improved |
| sample bridge index a from beta distribution instead of uniform (ablation) | DAVI · FFHQ Gaussian deblur | FID: 47.41 -> 34.34 | Improved |
| use monotonically increasing perturbation schedule instead of constant (ablation) | DAVI · FFHQ Gaussian deblur | FID: 34.34 -> 29.92 | Improved |
| single-step amortized inference instead of iterative sampling | inverse problem solvers · FFHQ 4x super-resolution | sec/image: 1.7 -> 0.04 (vs DDRM) | Improved |
| use DAVI instead of noise-scale-dependent solvers when sigma_y is unknown | Gaussian deblur · ImageNet 256 | FID: 157.38 -> 63.37 (vs DDRM) | Improved |
| use DAVI instead of RED-diff on Gaussian deblur | inverse problem solvers · FFHQ 256 | PSNR: 26.44 -> 25.46 | Mixed |
| use DAVI instead of DiffPIR on Gaussian deblur | inverse problem solvers · ImageNet 256 | FID: 62.48 -> 63.29 | No clear effect |
Closest papers by meaning
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.93)
- Divide-and-Conquer Posterior Sampling for Denoising Diffusion priors (NeurIPS 2024 · similarity 0.92)
- Improving Diffusion Inverse Problem Solving with Decoupled Noise Annealing (CVPR 2025 · similarity 0.91)
- A Mixture-Based Framework for Guiding Diffusion Models (ICML 2025 · similarity 0.90)
- Hierarchical Variational Policies for Reward-Guided Diffusion (NeurIPS 2026 · similarity 0.90)
- Adapt and Diffuse: Sample-adaptive Reconstruction via Latent Diffusion Models (ICML 2024 · similarity 0.90)
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.
