|
Sean ManI am a PhD student at the Technion, where I conduct research in the GIP Lab in the Computer Science Department and am enrolled in the Technion Autonomous Systems Program (TASP), under the supervision of Prof. Miki Elad. My research focuses on solving image inverse problems under realistic settings, such as unknown degradations and latent priors, with an emphasis on high perceptual quality. I was an Applied Scientist Intern at Amazon AWS, where I worked on masked discrete diffusion models for vision–language tasks. I have also served as a teaching assistant for Numerical Algorithms for six semesters, receiving Outstanding Teaching Assistant recognition for the last three consecutive times. Email / GitHub / Google Scholar / LinkedIn |
ResearchI'm interested in computer vision, machine learning, optimization, graphics and visual effects. |
|
DODO: Discrete OCR Diffusion Models, Roy Ganz, Roi Ronen, Shahar Tsiper, Shai Mazor, Niv Nayman arXiv, 2026 arxiv • The first VLM to use block discrete diffusion, enabling parallel block-wise decoding OCR that matches strong autoregressive baselines while delivering substantial throughput gains. |
|
ELAD: Blind Face Restoration using Expectation-based Likelihood Approximation and Diffusion Prior, Guy Ohayon, Ron Raphaeli, Matan Kleiner, Michael Elad SIGGRAPH Asia, 2025 arxiv • acm • website • We propose a novel approach to blind face restoration using a degradation estimator, expectation-based likelihood approximation, and diffusion prior. |
|
SILO: Solving inverse problems with latent operatorsRon Raphaeli, , Michael Elad ICCV, 2025 arxiv • cvf • talk • code • website • We propose a novel approach to solving inverse problems with latent diffusion models by training an extremely small latent operator to mimic the degradation operator. |
|
High-perceptual quality JPEG decoding via posterior sampling 🏅, Guy Ohayon, Theo Adrai, Michael Elad CVPRW, 2023 Sony Best Paper Award arxiv • cvf • We propose a novel Bayesian approach to JPEG decoding that leverages posterior sampling to improve the perceptual quality of the decoded image. |
|
Design and source code from Jon Barron's website and Leonid Keselman |