Lei Zhang
Senior Algorithm Engineer | Multi-modal Retrieval Researcher
Professional Summary
Experienced algorithm engineer with 8+ years of expertise in computer vision and multi-modal retrieval. Specialized in Person Re-Identification (ReID), Vision-Language Models (VLM), and large-scale image retrieval systems. Proven track record of leading technical teams to achieve world-record performance on international benchmarks and winning multiple 1st place awards in national AI competitions. Currently focusing on cutting-edge research in CLIP, VLM2Vec, and multi-modal representation learning at SenseTime.
Research Interests
Professional Experience
SenseTime Research
Apr 2022 – Present
R&D Department | Senior Researcher
Leading research in multi-modal retrieval, including CLIP, Vision-Language Models (VLM),
and VLM2Vec. Focus on applying vision-language pre-trained models to retrieval tasks and optimizing
multi-modal technologies for real-world applications. Driving innovation in cross-modal embedding learning
and large-scale retrieval systems.
Dahua Technology - Advanced Technology Research Institute
May 2018 – Apr 2022
AI Algorithm Department | Technical Lead (7-person team)
Served as Head of Human/Vehicle Feature Engineering and
Leader of Image-to-Image Search Team (7 engineers).
Responsible for feature performance optimization, retrieval system design and deployment.
Led the team to achieve 1st place in multiple major project competitions
and national AI challenges. Successfully deployed image search systems in various smart city projects.
Uniview Technology
Jul 2016 – May 2018
Intelligent Algorithm Department | Algorithm Engineer
Conducted R&D on backend server intelligence algorithms, focusing on
person retrieval and person attribute recognition.
Responsible for algorithm engineering deployment and performance optimization,
transforming research outcomes into deployable high-performance production solutions.
Competitions & Awards
5+
National Competition Awards
3
Years World Record Holder
4
First Place Wins
-
Market1501 / DukeMTMC-reID / CUHK03 - Three Major Benchmark DatasetsBroke world records for three consecutive years on all three major person Re-ID benchmarks
-
PRCV 2018 Large-Scale Person Retrieval Challenge1st Place in both Image Retrieval Track and System Performance Track
-
National Artificial Intelligence Competition - Person Re-ID TrackOrganized by Peng Cheng Laboratory | National Class-A Competition | Final Excellence Award
-
Ministry of Public Security - Vehicle Image Recognition AI Algorithm CompetitionOrganized by Traffic Management Research Institute of MPS | Authoritative vehicle Re-ID competition
-
Complex Environment Ground Target Detection Challenge - Long-Range Person IdentificationChallenging long-range person re-identification in complex scenarios
Publications
-
Paper Link: https://scholar.google.com/citations?user=icTGm1IAAAAJ&hl=zh-CN