Zhiyuan Cheng
Zhiyuan Cheng has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Chang Tan1 shared publication
- Huijie Liu
School of Artificial Intelligence and Data Science, University of Science and Technology of China, Hefei, China
1 shared publication - Jingmin Pan
Anhui Institute of Information Technology, China
1 shared publication - Jinze Wu
FLYTEK AI Research, Hefei, China
1 shared publication - Jun Yu
USTC iFLYTEK Co., Ltd., Hefei 230088, China
1 shared publication - Qi Liu
University of Science and Technology of China
1 shared publication
Latest Publications
- 28 January 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 311
EduYOLO: A classroom behavior recognition framework based on high-resolution feature attention fusion
- 28 January 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 311
EduYOLO: A classroom behavior recognition framework based on high-resolution feature attention fusion
- Open Access
- 22 March 2024
- Published by Association for Computing Machinery (ACM)
- p. 150-157
C2F-YOLO: A Coarse-to-Fine Object Detection Framework Based on YOLO
Benefiting from the development of deep learning in recent years, there have been significant advancements in object detection. In smart classroom environments, student behaviors vary widely due to differences in layout, lighting, and other factors unique to each smart classroom. This variability often leads to behavior confusion when using object detection algorithms to recognize student actions, which has somewhat hindered the progress of object detection in the field of smart education. Therefore, addressing this issue is both challenging and meaningful. To tackle this problem, this paper proposes a novel framework called C2F-YOLO to eliminate ambiguities in discerning student behaviors, thereby enhancing the performance of coarse-to-fine object detection tasks in smart education settings. Specifically, our framework consists of three main parts. Firstly, we introduce a novel method about coarse-to-fine object detection framework based on YOLO, C2F-YOLO, which identifies and refines ambiguous student behaviors in a coarse-to-fine manner. Secondly, we propose the cross-stage connection module named CSCM and the multi-scale feature fusion module named FSFM. The CSCM is used to transfer features of different scales from the first stage to the second stage to learn more diverse dimensional features, while the FSFM is used to merge features of different scales from the first stages. We conducted extensive experiments on our proprietary real-world dataset to demonstrate the superiority of our proposed framework. Remarkably, we achieved impressive results without significantly increasing the parameter count.
- 22 March 2024
- Published by Association for Computing Machinery (ACM)
- p. 150-157
Open AccessC2F-YOLO: A Coarse-to-Fine Object Detection Framework Based on YOLO
Benefiting from the development of deep learning in recent years, there have been significant advancements in object detection. In smart classroom environments, student behaviors vary widely due to differences in layout, lighting, and other factors unique to each smart classroom. This variability often leads to behavior confusion when using object detection algorithms to recognize student actions, which has somewhat hindered the progress of object detection in the field of smart education. Therefore, addressing this issue is both challenging and meaningful. To tackle this problem, this paper proposes a novel framework called C2F-YOLO to eliminate ambiguities in discerning student behaviors, thereby enhancing the performance of coarse-to-fine object detection tasks in smart education settings. Specifically, our framework consists of three main parts. Firstly, we introduce a novel method about coarse-to-fine object detection framework based on YOLO, C2F-YOLO, which identifies and refines ambiguous student behaviors in a coarse-to-fine manner. Secondly, we propose the cross-stage connection module named CSCM and the multi-scale feature fusion module named FSFM. The CSCM is used to transfer features of different scales from the first stage to the second stage to learn more diverse dimensional features, while the FSFM is used to merge features of different scales from the first stages. We conducted extensive experiments on our proprietary real-world dataset to demonstrate the superiority of our proposed framework. Remarkably, we achieved impressive results without significantly increasing the parameter count.