Haiyi Mei 梅海艺
Haiyi Mei (梅海艺) is an engineer, researcher and pianist, and a nomad living in Chiang Mai, Thailand for now. Haiyi works at EmotionX, an emotional-technology company. Before that, Haiyi designed and led the core agentic framework of CREAO AI's coding platform, and at SenseTime built synthetic-data systems and worked on 3D human research and video generation. On the side Haiyi makes ScoreStarling, CinemaCC and Chord Spell.
Side projects
- ScoreStarling: Turns recordings into editable scores, for one instrument, both hands at the piano or a whole band with its own parts, drums included; or start from a PDF or MIDI, then edit, listen and export.
- CinemaCC: A local-first subtitle companion that keeps one or two SRT tracks synchronized beside a television, laptop, projector, or cinema screen.
- CinemaCC Subtitle Skill: An open, auditable workflow for researching, repairing, translating, and validating movie or TV subtitles without silently changing cue timing.
- Chord Spell: An iOS app for practising how chords are spelled: a chord name appears, and you tap its notes on a piano keyboard.
Work
- EmotionX: where Haiyi works now, on emotional technology.
- R&D Engineer, Creao AI, Apr 2025 - Mar 2026: Designed and led the core agentic framework for a production coding platform that scaled beyond 100,000 registered users.
- Algorithm Researcher, SenseTime, June 2021 - Apr 2025: Built text-to-video and synthetic-data systems, from VAE training and automated annotation to scalable rendering infrastructure.
- Visiting Research Intern, NLPR | CASIA, 2018 - 2021: Started in synthetic data and computer vision, including video captioning research and early rendering-system experiments.
Research
- SMPLest-X: Ultimate Scaling for Expressive Human Pose and Shape Estimation, TPAMI 2025
- Differentiable Convex Polyhedra Optimization from Multi-view Images, ECCV 2024
- WHAC: World-grounded Humans and Cameras, ECCV 2024
- Digital Life Project: Autonomous 3D Characters with Social Intelligence, CVPR 2024
- AiOS: All-in-One-Stage 3D Wholebody Mesh Recovery, CVPR 2024
- PrimDiffusion: Volumetric Primitives Diffusion for 3D Human Generation, NeurIPS 2023
- SMPLer-X: Advanced 3D Human Body Modeling, NeurIPS 2023
- SynBody: Synthetic Dataset with Layered Human Models, ICCV 2023, co-first author
- Zolly: Zoom Focal Length Correctly, ICCV 2023 (Oral)
- SHERF: Generalizable Human NeRF from a Single Image, ICCV 2023
- HumanLiff: Layer-wise 3D Human Generation with Diffusion Model, Preprint
Email · LinkedIn · GitHub · Google Scholar · YouTube · Bilibili · Nomads