電腦視覺為本實驗室的核心研究領域，我們專注於跨攝影機、跨場域與多源資料下的物件偵測挑戰。研究整合聯邦學習、無監督/半監督領域自適應與圖像生成技術，以提升模型在跨領域資料中的泛化能力。

Computer vision is the core focus of our lab, with a particular emphasis on object detection across cameras, domains, and heterogeneous data sources. Our research integrates federated learning, unsupervised and semi-supervised domain adaptation, and image generation techniques to enhance model generalization under domain shifts.

透過結合視覺、語言等多模態資訊，本研究方向深入探討視覺語言模型（VLM）的限制，以及大語言模型與視覺模型之間的對齊機制，以突破模型在開放環境中的應用瓶頸。

We focus on multimodal learning, with an emphasis on open-vocabulary understanding and semantic reasoning to enhance models' ability to recognize novel categories. Our research investigates the limitations of vision-language models (VLMs) and explores alignment mechanisms between large language models (LLMs) and visual encoders to address challenges in real-world scenarios.

本實驗室致力於將最新電腦視覺、深度學習與大數據分析技術應用於實務場域，包括環境異常偵測、環境犯罪調查、跨攝影機可疑車輛與行人偵測，以及智慧督查等應用場景。

We apply the latest technologies in computer vision, deep learning, and big data analysis to practical scenarios, including environmental anomaly detection, environmental crime investigation, cross-camera detection of suspicious vehicles and pedestrians, and intelligent inspection tasks in industrial and public domains.