I build computer vision pipelines and C++ applications, with a focus on measured results and runnable demos.
A fixed-camera parking occupancy pipeline that maps 553 spaces, trains a compact classifier in Python, and processes video through a C++17 application using OpenCV and ONNX Runtime. The classifier reached 97.56% accuracy on 164 test crops from the same camera domain. The repository includes evaluation figures, an annotated video, and a complete C++ video inference path.
A ResNet18 traffic-sign classifier with training, evaluation, ONNX export, and C++ inference. It reached 99.04% accuracy on the official 12,630-image test set. The C++ application uses OpenCV preprocessing and ONNX Runtime, with checks comparing its outputs with the PyTorch pipeline.
A C++17 / Qt Widgets desktop application for simulated temperature and humidity monitoring. It provides threshold alarms, SQLite history, data restoration across launches, and Qt Test checks. The project uses sliders in place of physical sensors.
- THUCNews BiLSTM Classification — Chinese news-title classification with a reproducible evaluation pipeline.
- Flowers102 ResNet18 — comparison of transfer learning and training from scratch.
- Qt Widgets Practice — an archive of three earlier Qt desktop exercises, with source and verification notes.
C++17 · Qt Widgets · OpenCV · ONNX Runtime · Python · PyTorch · TensorFlow · SQLite · CMake