Real-Time Edge AI Optimization on Embedded SoC
Description
Optimized deep learning models for real-time inference on embedded hardware using TensorRT and CUDA. SOTA performance and memory usage through FP16/INT8 quantization, kernel fusion, and profiling-driven optimization.
Period
At stradvision (real), and 2025.12 ~ current (toy)
Skills
C++, CUDA, TensorRT, ONNX, Linux, Docker, GStreamer, OpenVX(TIOVX), OpenCV, Python
Dataset-Free Knowledge Distillation with Synthetic Data
Description
Implemented a dataset-free knowledge distillation framework leveraging synthetic data generation and contrastive feature learning. Achieved reduced student model training time while maintaining competitive accuracy.
Period
March 2026 at Korea University
Skills
Python, PyTorch, CUDA, NumPy, Matplotlib
Surround View Monitor (SVM) – 360° Multi-Camera Embedded System
Description
designed and developed a real-time surround view monitor system integrating four automotive cameras to generate a stitched bird's-eye view. Implemented camera calibration, geometric transformation, and low-latency image stitching optimized for embedded hardware deployment.
Period
~ 2019 at NEXTCHIP
Skills
C++, OpenCV, CUDA, Embedded Linux, GStreamer, EGL/OpenGL ES, Multi-threading
Dedicated Embedded File System for DVR Recording System
Description
Designed and implemented a custom file system optimized for continuous high-throughput video recording in DVR systems. Focused on minimizing fragmentation, ensuring crash recovery robustness, and supporting cyclic overwrite storage for long-duration surveillance recording.
Period
2007 ~ 2011.03 at ITX Security
Skills
C, C++, Embedded Linux, Flash Storage Management, Multi-threading, Low-level I/O (Scatter-gather), Filesystem Design (FAT32 compatible)
