Bin Sun
Research Scientist at Adobe
Research
Product
Deployment
Full-stack AI researcher across research, product, and deployment
Generative computer vision · Production models · Research-driven startups
Adobe · Dallas, TX · Ph.D., Northeastern University
About
I am a full-stack AI researcher spanning research, product, and deployment. I publish and invent in generative computer vision, turn research into shipped product capabilities, and own the path from model idea to production system—rather than specializing in only one stage of the stack.
I am a Research Scientist at Adobe, following Adobe’s September 2026 acquisition of Topaz Labs. My recent product work includes Wonder 4, Wonder 3 / 3.5, and Recover v3—covering research, model development, productization, and production deployment for generative image enhancement and restoration. Wonder 3.5 has accounted for more than half of the company’s image-model revenue, and Wonder 4 is the newest flagship: an all-in-one mixture-of-experts model that upscales and repairs images in a single pass. Earlier, I led diffusion-model product research at Picsart AI Research, and at Topaz I took generative image-editing research into a shipped product feature.
I also bring a startup-builder perspective. I was a founding member of Giaran, where I helped build real-time facial-analysis technology before the company was acquired by Shiseido, and I later worked with AInnovationLabs. Across these roles, I have repeatedly worked in small, research-driven teams where technical ideas had to become usable products.
I received my Ph.D. in Electrical and Computer Engineering from Northeastern University, advised by Prof. Yun Fu at SMILE Lab. My research has appeared at ICLR, CVPR, AAAI, TNNLS, BMVC, and related venues.
I am a Research Scientist at Adobe, following Adobe’s September 2026 acquisition of Topaz Labs. My recent product work includes Wonder 4, Wonder 3 / 3.5, and Recover v3—covering research, model development, productization, and production deployment for generative image enhancement and restoration. Wonder 3.5 has accounted for more than half of the company’s image-model revenue, and Wonder 4 is the newest flagship: an all-in-one mixture-of-experts model that upscales and repairs images in a single pass. Earlier, I led diffusion-model product research at Picsart AI Research, and at Topaz I took generative image-editing research into a shipped product feature.
I also bring a startup-builder perspective. I was a founding member of Giaran, where I helped build real-time facial-analysis technology before the company was acquired by Shiseido, and I later worked with AInnovationLabs. Across these roles, I have repeatedly worked in small, research-driven teams where technical ideas had to become usable products.
I received my Ph.D. in Electrical and Computer Engineering from Northeastern University, advised by Prof. Yun Fu at SMILE Lab. My research has appeared at ICLR, CVPR, AAAI, TNNLS, BMVC, and related venues.
1,300+
reported research citations*
11
h-index
440+
citations on co-first-author challenge work
59
completed peer reviews (reported)
*Self-reported citation count as of September 2026; includes publications and patents. See Google Scholar for the current indexed count.
Research × Product × Deployment
Research
Novel methods with external validation
1,300+ reported research citations, ICLR/CVPR Oral papers, AAAI and TNNLS publications, patented technologies, and a CVPR challenge-winning co-first-author work.
Product
Research that becomes shipped capability
Owned generative image-editing, restoration, and enhancement models through productization—including Generative Remove, Recover v3, Wonder 3 / 3.5, and Wonder 4. Wonder 3.5 alone has driven more than half of Topaz image-model revenue.
Deployment
From training pipeline to production systems
End-to-end ownership across architecture, training, evaluation, optimization, and production release—so models leave the lab as reliable systems users can depend on.
Positioning: a full-stack AI researcher who connects research breakthroughs, product decisions, and production deployment into one continuous loop.
Experience
Research Scientist · Adobe Topaz Labs acquisition
Sep 2026 – Present
Previously Research Scientist at Topaz Labs (2025 – Sep 2026)
Research, productization, and deployment of generative image models
- Co-developed Wonder 4, Topaz’s newest flagship all-in-one image model: mixture-of-experts architecture that combines upscaling, denoising, deblurring, compression-artifact removal, and face recovery in a single pass.
- Led research and core model development for Wonder 3 and Wonder 3.5; Wonder 3.5 has accounted for more than half of company image-model revenue.
- Developed Recover v3 for restoration of degraded and low-quality imagery.
- Owned the full stack from research direction through training, evaluation, product integration, and production deployment.
Senior Research Scientist · Picsart AI Research
2024 – 2025
Generative AI research and productization
- Led research on Stable Diffusion-based models for high-quality image generation and editing.
- Drove the research-to-product path: model design, training, evaluation, and productization for production generative-AI capabilities.
Deep Learning Researcher · Topaz Labs
2023 – 2024
Generative editing research shipped as product
- Developed generative inpainting research into a shipped product feature: Generative Remove.
- Bridged research and deployment by adapting Stable Diffusion-style models for reliable product release on consumer hardware.
Startup & Early Industry Experience
2017 – 2022
- Founding member, Giaran; helped build the core real-time facial alignment / tracking technology and experienced the startup journey through acquisition by Shiseido.
- Computer Vision Engineer, Zebra Technologies; developed automated chute-fullness detection technology.
- Worked with AInnovationLabs on research-driven startup development, expanding experience across early-stage AI product building.
Selected Highlights
- November 2026: Wonder 4 launched on Topaz for Web / Mobile as the new flagship all-in-one image model (mixture-of-experts); co-authored the technical report Wonder 4: A Mixture-of-Experts Recipe for Image Quality Enhancement.
- September 2026: Adobe completed its acquisition of Topaz Labs; continue research-to-product work on generative image models.
- August 2026: Wonder 3.5 launched; led core model research, productization, and deployment ownership. Wonder 3.5 has accounted for more than half of company image-model revenue.
- 2025–2026: Shipped production image models including Recover v3, Wonder 3 / 3.5, and Wonder 4 at Topaz Labs / Adobe.
- 2024: Joined Picsart AI Research as Senior Research Scientist and led Stable Diffusion-based product research.
- 2023: Took generative image-editing research into a shipped Topaz product feature, connecting research, product, and deployment.
- 2023: Image as Set of Points accepted to ICLR as a top-5% Oral.
- 2022: Towards Layer-wise Image Vectorization accepted to CVPR as an Oral.
- 2021: Team won 1st place in both RGB and RGB-D tracks of the CVPR Sign Language Recognition Challenge.
- 2017: Giaran, where I was a founding team member, was acquired by Shiseido.
Selected Publications [Full list on Google Scholar]
Patents & Technology
- US20190014884A1 — Systems and Methods for Virtual Facial Makeup Removal and Simulation, Fast Facial Detection and Landmark Tracking, Reduction in Input Video Lag and Shaking, and Makeup Recommendation (Shiseido)
- WO2019213459A1 — System and Method for Generating Image Landmarks
- US11210549B2 — Automated Chute Fullness Detection (Zebra Technologies)
- WO2020247545A1 — Light-Weight Decompositional Convolution Neural Network
- WO2021163103A1 — Light-Weight Pose Estimation Network with Multi-Scale Heatmap Fusion
- US20230153946A1 — System and Method for Image Super-Resolution
Awards & Professional Service
- 1st Prize in both RGB and RGB-D tracks, CVPR 2021 Challenge on Large Scale Signer Independent Isolated Sign Language Recognition.
- 4th Place, CVPR 2021 Challenge on Agriculture Vision.
- Star Intern of the Year, Zebra Technologies, 2019.
- GapFund360 Award, Northeastern University, 2018.
- Peer review: Approximately 59 completed reviews (self-reported) across conferences and journals including CVPR, ICCV, ICLR, AAAI, IJCAI, TPAMI, TIP, TCSVT, and other venues.