Ly Vireakdara

I’m a computer vision researcher specializing in open-vocabulary object detection and multi-object tracking robustness. I’m the first author on EnYOLO-World, where I improved the YOLO-World architecture with PGI and FiLM-PAN, delivering +3.1 AP (52.2 AP on COCO) at 78 FPS (under review at ELCVIA), and I lead VocabDriftMOT, the first benchmark for vocabulary-drift failure modes in OV-MOT (510 annotated events, 21 sequences).

Alongside research, I bring 4 years of production software engineering experience, including full-stack delivery and cloud deployment. I’ve built and maintained systems serving 10,000+ users, designed PostgreSQL schemas and optimized queries (cutting API response times by 40%), and implemented high-throughput REST APIs with 99.9% uptime—so my work ships from model idea to reliable service.