About Me
I am Duy Khanh Dinh, a Computer Vision Team Leader at VNPT AI (VNPT Group). My primary focus lies in Computer Vision, Generative Models (Diffusion models, FLUX architectures), and Biometric Security (Presentation Attack Detection & Injection Attack Detection).
Previously, I worked as a Research Assistant at the Singapore University of Technology and Design (SUTD) on the CISDeM project, focusing on machine learning for ransomware detection and encrypted malware traffic analysis. Earlier in my career, I served as a System Software Engineer and Power Optimization Team Leader at Samsung Vietnam Mobile R&D Center (SVMC) working on Linux device drivers and the Tizen OS framework.
I earned my Professional Master's Degree in Data Science and AI from the Asian Institute of Technology (AIT) (Valedictorian, GPA 3.94/4, Rank 1/14) and my Engineering Degree in Electronics and Communication Engineering from Hanoi University of Science and Technology (HUST) (Rank 8/600).
Research Interests
- Computer Vision & Deep Learning: Object detection, video anomaly detection and understanding, multimodal reasoning.
- Biometric Security: Face Anti-Spoofing (iBeta Level 1 & 2 compliance), Injection Attack Detection, 1:1 and 1:N Face Recognition.
- Generative Models: Diffusion models, concept unlearning, personalized text-to-image synthesis.
Recent News & Highlights
- [WACV 2027]: Our paper “Unlearn in Context: Contrastive Alignment via Diversified Supervision for Diffusion Models” has been accepted to WACV 2027.
- [ECCV 2026]: Ranked Top 4 on the Private Leaderboard in the AI City Challenge Track 3 (Anomalous Events in Transportation).
- [ICCV 2025 & CVPR 2024]: Ranked Top 1 on the Private Leaderboard in the AI City Challenge Track 4 (Fisheye Road Object Detection).
- [CVPR 2024 & 2023]: Ranked Top 3 and Top 4 on the Private Leaderboard in the AI City Challenge Track 5 (Motorcyclist Helmet Violation Detection).
- Biometric Security Certification: Successfully achieved international iBeta Level 1 & Level 2 certifications for VNPT’s Face Anti-Spoofing SDK.
