雷晨阳
Since 2017, I’ve built visual systems that bridge what a camera captures and what a person actually wants—from computational imaging to multimodal AI, with work deployed on flagship phones and released as open-source models.
Currently: Technical Expert, Celia Lab, Huawei
At Huawei’s Celia Lab, I previously served as technical lead for Celia AI Photo Editing (一句话P图), a flagship AI experience showcased at the HUAWEI Mate 80 series launch that has surpassed 100M uses. I also initiated Project Boogu and led its first open-source release, Boogu-Image-0.1—a 10B Apache-2.0 model family trained on 208.62M unique images at an estimated base-model cost of about $400K. I now lead the team building Boogu’s next generation across visual understanding, generation, and editing.
Previously: Assistant Professor at CAIR/HKISI, Chinese Academy of Sciences (2022–2025); Visiting Scholar at Princeton with Felix Heide (2022–2024), on computational imaging. Ph.D. from HKUST in 2022 with Qifeng Chen, B.Eng. from Zhejiang University in 2018. Research internships at MSRA, NVIDIA, and SenseTime.
Research Summary
The recurring question in my work is: what information is missing between a physical scene, a captured signal, and the image a person intends?
I approach this as a systems problem. Depending on where the information is lost, I change the sensing process, the reconstruction method, or the multimodal model — not just the network scale. Earlier, this meant polarization-aware sensing and temporally coherent reconstruction; more recently, multimodal understanding, generation, and editing.
All of these recover what was missing after the fact. I am now interested in the return leg — deciding how a photograph should be taken in the first place: reasoning jointly about what makes an image good, the physical structure of the scene, and what a real camera can actually do.
From Intent to Final Image
I aim to build autonomous photography systems that translate creative intent into final images by jointly reasoning about visual aesthetics, physical environments, camera capabilities, and post-processing. Such systems could perceive and act in the real world, plan and execute shots, and collaborate with people across smartphones, intelligent cameras, drones, and robotic imaging platforms.
Capture richer physical signals through polarization-aware sensing and sensor–algorithm co-design.
Brings shape from polarization to complex real-world scenes, with the first scene-level dataset and single-image surface-normal reconstruction.
CVPR 2024
Polarization Wavefront LiDAR
Goes beyond depth-only LiDAR by measuring polarized wavefronts to jointly recover distance, surface normals, and material properties, demonstrated with a 223-meter eye-safe hardware prototype.
Reconstruct coherent images and videos from incomplete, degraded, or temporally inconsistent observations.
Uses the internal statistics of a target video to make frame-wise image processing temporally coherent.
CVPR 2021 · TPAMI 2023
Flash-assisted Reflection Removal
Introduces a reflection-free flash-only cue in RAW space, improving reflection removal by more than 5 dB over prior methods.
Translate visual content and human intent into generation and editing systems under real training and deployment constraints.
A flagship AI editing experience showcased at the HUAWEI Mate 80 series launch and deployed at exceptional scale, surpassing 100M uses.
The first release of Project Boogu: a 10B Apache-2.0 model family unifying image generation and editing. Trained on 208.62M unique images at an estimated base-model cost of about $400K, it matches or surpasses leading open models across standard benchmarks and approaches top closed systems.
Bilibili reviewAI WeeklyComfyUI guide
Reason about image quality, scene structure, and camera capability in the physical world, before the shutter is pressed.
Learns where, what, and how to shoot by translating visual intent into coordinated decisions about viewpoint, composition, subject arrangement, and camera configuration.
Research Prototype · 2024
Vision-Language-Camera
Connects photographic intent with professional camera control through vision-language reasoning over image appearance, camera parameters, and actionable shooting guidance.
| Institution | Role | Period |
|---|---|---|
| Huawei, Celia Lab | Technical ExpertProject Leader of Boogu-ImageTechnical Lead for “一句话P图” | 2025 – Present |
| CAIR, HKISI, Chinese Academy of Sciences | Assistant Professor | 2022 – 2025 |
| Princeton University | Visiting Scholar | 2022 – 2024 |
| HKUST | Ph.D. in Computer ScienceAdvised by Qifeng ChenRedBird PhD Scholarship, 2021 | 2018 – 2022 |
| NVIDIA | Research Intern | 2022 |
| Microsoft Research Asia (MSRA) | Research Intern | 2021 – 2022 |
| SenseTime | Research Intern | 2019 – 2020 |
| Zhejiang University | B.Eng.Outstanding Graduate, 2018National Scholarship, 2017 | 2014 – 2018 |
Academic Service
HKUST