ICIP 2026 plenary: computational imaging shifts from explicit priors to learned operators
prof_kamilov · x · 2026-09-15
Ulugbek Kamilov delivers an ICIP 2026 plenary (Sept 13-17, Tampere, Finland) arguing that generative modeling is driving a fundamental shift in computational imaging: from explicitly defined priors to learned restoration operators that implicitly encode data distributions.
Key points:
- Classical inverse-problem formulations rely on explicit priors — principled but limited for real-world data
- Priors are now accessed via learned models that can be queried but not written down, reframing inverse problems as iterative algorithms interacting with these operators
- Three ideas: general restoration operators implicitly define priors; stochasticity is a principled tool for highly ill-posed problems; priors can be learned directly from measurements without clean training data
The talk connects proximal methods, plug-and-play algorithms, and modern generative models, with biomedical image reconstruction applications.
More from Research
- TailSFT: Skipping SFT Gains Improves pass@16 and RL Exploration in GRPO — tw_killian · 2026-09-15
- STARK Co-inventor Notes STARKs Are Post-Quantum Secure — jamestagg · 2026-09-15
- Self-evolving AI agents shouldn't grade their own homework: a 7-step evidence-gated roadmap — MaryamMiradi · 2026-09-15
- I2T loss stays comparable across tokenizers and predicts image generation quality — peterxichen · 2026-09-15
- Fruit fly brain with 166,700 neurons installed in a physical robot that walks — udmrzn · 2026-09-15
- Image Tokenizers Define the Visual Language of Unified Multimodal Models — peterxichen · 2026-09-15