My Background
I am a Masters student at Universiti Teknologi Malaysia, researching active learning for object detectors.
My Experience
I have been in the video analytics industry for almost two years, building and optimizing computer vision models and pipelines for production.
Algorithm Engineer (Computer Vision) @ Graymatics (Sep 2023 - Aug 2024)
- Implemented batching and motion-based frame skipping in DeepStream to optimize inference server’s GPU utilization and memory usage, improving FPS for real-time analytics involving 140+ cameras.
- Created a lightweight skeleton-based action recognition architecture, achieving ~90% accuracy at 500+ FPS.
- Resolved a critical bug affecting server stability and improved multi-GPU load distribution to achieve an inference throughput of upto 300 streams per server.
Computer Vision Consultant @ Innotec Solutions (Jul 2023 - Aug 2023)
- Provided expert advice on model training and deployment queries, enabling them to build in-house video analytics solutions.
Computer Vision Intern @ Tapway (Jan 2022 - Sep 2022)
An AI startup where I worked on projects from high-profile clients such as PLUS, RTS, KLK and others which are now deployed to production.
Python Development
- Developed an efficient inference pipeline to detect and track anomalous objects with ~90% accuracy.
- Implemented intelligent runtime ROI adjustment to tackle continuous variations in ROI, improving counting accuracy.
Model Inference Optimization
- Reduced resource starvation and latency spikes in multi-stream inference by optimizing the number of threads spawned by each process.
- Fixed a major memory leak in the inference pipeline that was causing noticeable degradation in performance.