Distilling One- Stage Object Detection Network via Decoupling Foreground and Background Features
學年 111
學期 1
發表日期 2022-10-24
作品名稱 Distilling One- Stage Object Detection Network via Decoupling Foreground and Background Features
作品名稱（其他語言）
著者 San-Hao Tsai; Shwu-Huey Yen; Hwei-Jen Lin
作品所屬單位
出版者
會議名稱 CICET'22: International Conference on Recent Advancements in Computing in AI, IoT and Computer Engineering Technology
會議地點 新北市，台灣
摘要 To deploy an object detection system into an application, the size of the system is one of key issues. We distill a teacher’s knowledge into our small scale network, where the teacher is a full-scale architecture with good performance. In addition to KL divergence loss, we propose a cosine similarity loss on foreground features to encourage student to learn the feature direction of teacher’s. This leads an efficient and robust learning from teacher model. We also propose an adaptive learning criteria which makes student model learns from teacher only when teacher has a better performance than student’s. The proposed student model has an improvement of 34.85% on ResNet34 and 39.67% on ResNet18 when Teacher model is on ResNet50.
關鍵字 Knowledge Distillation;Decoupling Features
語言 en
收錄於
會議性質 國際
校內研討會地點 無
研討會時間 20221024~20221026
通訊作者
國別 TWN
公開徵稿
出版型式
出處 Proceedings of CICET 2022
相關連結

機構典藏連結 ( http://tkuir.lib.tku.edu.tw:8080/dspace/handle/987654321/123442 )