演講時間: 113年5月15日（三） 14:00~16:00

演講地點: E6-A207教室

演講者: 盧沛怡助理教授（國立中正大學資工系）

演講主題: Federated Learning-based Multi-source Domain Adaptation for Object Detection

大 綱： A new research topic, privacy-preserving multi-source domain adaptation for object detection is explored. To ensure data privacy, we adopt federated learning framework as the fundamental architecture, which contains multiple clients (source domains) and a single server (target domain). Accordingly, we employ some domain adaptation techniques and source-only methods that can be applied to clients, as well as model aggregation algorithms that can be adopted on the server. Subsequently, we propose various frameworks that utilize the above methods on clients and the server, respectively. To evaluate the effectiveness of our proposed methods, two benchmark scenarios are adopted in this research, including nighttime pedestrian detection such and commonly used domain adaptation dataset. Experimental results demonstrate that while preserving data privacy, our methods can outperform the state-of-the-art multi-source domain adaptive object detection methods for almost all scenarios of interest.

請研究所一年級的同學當天準時聽講