# Chengqi (William) Li

> Personal homepage of Chengqi (William) Li, focusing on computer vision for robotics and autonomous driving, with projects in 3D perception, traffic anomaly detection, and model deployment.

## About

- [Homepage](https://chengqili.com/): Bio, research interests, and selected projects.

## Projects

- [Image-Pretrained Vision Transformers for Real-Time Traffic Anomaly Detection](https://chengqili.com/projects/videomt_tad/): Real-time traffic anomaly detection with image-pretrained ViTs (78.1% AUC-ROC at 200 FPS on DoTA).
- [VRU Tracking in LiDAR Point Cloud](https://chengqili.com/projects/object_tracking/): 3D pedestrian tracking in LiDAR point clouds inspired by spatio-temporal tracking paradigms.
- [YourArXiv -- Personalized ArXiv Recommendation](https://chengqili.com/projects/youraxiv/): Embedding-based recommendation system for arXiv papers using Ada2 embeddings.
- [YOLOv5 Deployment and Optimization Pipeline](https://chengqili.com/projects/quantize/): 2-stage deployment pipeline using knowledge distillation and quantization for real-time inference.

## Blog Posts

- [How Transformer Works in Machine Translation](https://chengqili.com/post/transformer/): In-depth look at the Transformer architecture with PyTorch implementation.
- [How Neural Networks Memorize Data](https://chengqili.com/post/representation/): Exploration of neural network memorization using positional encoding and MLPs.