SGPnP paper accepted at IEEE TCI: diffusion priors plug into PnP reconstruction with theory
prof_kamilov · x · 2026-10-18
A team from UW-Madison, WashU, and Los Alamos National Laboratory published "Stochastic Generative Plug-and-Play Priors" (SGPnP), accepted by IEEE Transactions on Computational Imaging.
Key contributions:
- Score-based PnP: pretrained score-based diffusion models can be used directly as priors inside classical Plug-and-Play reconstruction algorithms, without reverse diffusion sampling
- Stochastic generative update: adding noise before denoising better matches the score model's training regime; the injected noise smooths the objective, helps avoid strict saddle points, and annealing yields convergence to a stationary point
- Theory: a score-based interpretation of PnP, compatible with ADMM, HQS, and PGM-style updates, with decoupled noise levels for injection and denoiser
- Results: improvements on hard inverse problems like large-mask inpainting and multi-coil MRI, beating standard PnP baselines and staying competitive with diffusion-based solvers
Paper and code are publicly available.
More from Research
- Work in progress: geometric structure of GPG projections as Markov chain coarsening — michaelchchoi · 2026-10-18
- Backprop-free pretraining gets 2.3x speedup as Noise method crushes Dust record on 8 GPUs — inductionheads · 2026-10-18
- q3dviewer: open-source Qt+OpenGL tool for painter-like point cloud editing — 4310sy · 2026-10-18
- Smarter Every Day's Destin to defend PhD dissertation on impact fusion — CatAstro_Piyush · 2026-10-18
- Brain implants are restoring speech to people with ALS, Science reports — EricMTrautmann · 2026-10-18
- Formal Verification of LLM-Generated Code: Slides from FATTW 2026 Talk — CatAstro_Piyush · 2026-10-18