Lei Sun (孙磊)

Research Scientist@INSAIT, adviced by Prof. Luc Van Gool and Dr. Danda Pani Paudel.

I earned my Ph.D. from Zhejiang University, privileged to be supervised by the best professor in the world——my life-long mentor and role model, Prof. Kaiwei Wang (汪凯巍). I also had the privilege to be supervised by Prof. Davide Scaramuzza at the Robotics and Perception Group, University of Zurich, and Prof. Luc Van Gool at the Computer Vision Lab, ETH Zurich. I also work closely with Prof. Ming-Hsuan Yang from UC Merced and Google DeepMind.

I'm interested in image/video generation, computational imaging, and event-based vision.

I am always open to collaborate. Drop me a email if you are interested!

Email  /  Google Scholar  /  Linkedin  /  Github

First-author papers and corresponding author papers are highlighted. * denotes equal contribution and † denotes corresponding author.

Removing lens veiling glare from images.

large-scale dataset and benchmark for photographic cameras, built via automatic optical design to cover diverse optical aberrations.

Better IQA metrics for diffusion-based SR models.

OmniLens, a flexible solution to universal computational aberration correction via large-scale LensLib pre-training and adapting the model to any specific lens designs with unknown lens descriptions via fast LensLib-to-specific domain adaptation.

We present a controllable human video generation framework that flexibly combines any subject, background, trajectory, and action to create realistic videos of anyone doing anything anywhere.

Best low-light image enhancement method with event camera so far.

The best algorithm (so far) for the automatic design of compound-lens-based computational imaging systems.

Unified Event-based frame interpolation for both sharp frames and blurry frames with unsupervised domain adaption for unkown camera parameters.

EvTemMap, pioneering work of event-to-image conversion, produces high-quality, high-grascale-resolution, high-dynamic-range image from events only.

Event-based depth estimation for dynamic scenes, over 41 times faster in data acquisition than previous SOTA.

A Panoramic Computational Imaging Engine to achieve minimalist and high-quality panoramic imaging.

Event-assisted fast auto-focus algorithm. Precise focus with less than one depth of focus is achieved within 0.004 seconds.

SOTA unified Event-based frame interpolation for both sharp frames and blurry frames.

We present a novel event-based deblurring method that improves the previous state-of-the-art by 2.47 dB.

Annular Computational Imaging (ACI) framework to break the optical limit of light-weight Panoramic Annular Lens design.

A lightweight event-based human pose estimation model by processing events as point cloud.

Designing a lightweight panoramic annular lens with physical-based image enhancement model.

Designing a panoramic annular lens (PAL) system with 4K high resolution for aerial image segmentation.

Designing a specific semantic segmentation for aerial panoramic images.

A real-time RGB-D fusion semantic segmentation framework with small obstacle detection.

Propose three different auditory-based interaction methods which convey raw depth images, obstacle information and path information respectively to visually impaired people.

Propose a panoramic localizer, which is based on coarse-to-fine descriptors, leveraging panoramas for omnidirectional perception and sufficient FoV up to 360◦.

Develop a wearable system to transform the spatial information captured by camera into a voice description and fed it back to blind users.

Propose a framework to alleviate the accuracy decline when semantic segmentation is taken to adverse conditions by using Generative Adversarial Networks (GANs).

Organizer/Program Chair of CVPR 2026 Workshop on Agentic AI for Visual Media.

Co-organizer of CVPR 2026 New Trends in Image Restoration and Enhancement (NTIRE) workshop.

Organizer of the Challenge On Event-based Image Deblurring, Challenge on Image Denoising, the First Blind Computational Aberration Correction Challenge, and the Efficient Super-Resolution Challenge

Co-organizer of CVPR 2025 New Trends in Image Restoration and Enhancement (NTIRE) workshop.

Organizer of the First Challenge On Event-based Image Deblurring in CVPR 2025 NTIRE workshop & Event-Based Vision workshop.

Organizer of the Challenge On Image Denoising.

Co-organizer of the Challenge On Image Super-Resolution, and Challenge on Real-World Face Restoration.

Area Chair for ICML 2026, NeurIPS 2026.

Reviewer for CVPR, ICCV, ECCV, AAAI, WACV, IJCV, TPAMI, TIP, RA-L, CVIU.

Dec. 2025: Invited talk on intellegent optical system in College of Optical Science and Engineering, Zhejiang University.

Dec. 2025: Invited talk on intellegent optical system in College of Computer Science, Chongqing University.

Oct. 2022: [ Video ] Invited talk on event-based image deblurring in Mobile Intelligent Photography and Imaging (MIPI) .

Jun. 2022: Invited talk in IV 2022 BSL Workshop.

Website template from Jon Barron