Illustration of the Overall Framework of MLFP
Abstract:
In the study of incremental object detection, knowledge distillation and data replay are effective methods to mitigate catastrophic forgetting. However, current research ...Show MoreMetadata
Abstract:
In the study of incremental object detection, knowledge distillation and data replay are effective methods to mitigate catastrophic forgetting. However, current research on single-stage detectors is limited, single-stage detector outputs often contain excessive negative sample information, and direct application of knowledge distillation to them is ineffective. To address this, this paper proposes a multi-level foreground prompt incremental learning algorithm for single-stage detectors like YOLO, including foreground prompts at the image level, feature map level, and knowledge level. First, to obtain fewer but more representative replay samples, representative images with a high number of old class foregrounds are selected by maximizing sample structure distance, providing direct foreground information at the image level. Second, the feature map output by the teacher model is used as a feature-level prompt, with a feature distillation loss guiding the student model to encode new class foreground information in less significant channels of the old feature map, reducing interference. Lastly, the teacher model’s inference output serves as a knowledge-level prompt, and the adaptive select object method is introduced to avoid foreground conflict in traditional knowledge distillation, enhancing the model’s plasticity by selectively merging foreground information. Extensive experiments on PASCAL VOC and MS COCO datasets demonstrate that this approach significantly improves model plasticity while maintaining stability.
Illustration of the Overall Framework of MLFP
Published in: IEEE Access ( Volume: 13)
Funding Agency:
School of Information and Communication, Guilin University of Electronic Technology, Guilin, China
Jianwen Mo received the B.S. degree from the Department of Electronic Engineering, North China Electric Power University, in 1994, the M.S. degree from the School of Mathematics and Computer Science, Guangxi Normal University, in 2002, and the Ph.D. degree from the School of Electronic Engineering, Xidian University, in 2011. He is currently working as a full-time Associate Professor with Guilin University of Electronic T...Show More
Jianwen Mo received the B.S. degree from the Department of Electronic Engineering, North China Electric Power University, in 1994, the M.S. degree from the School of Mathematics and Computer Science, Guangxi Normal University, in 2002, and the Ph.D. degree from the School of Electronic Engineering, Xidian University, in 2011. He is currently working as a full-time Associate Professor with Guilin University of Electronic T...View more
School of Information and Communication, Guilin University of Electronic Technology, Guilin, China
Ronghua Zou received the B.S. degree from Guangzhou Maritime University, China, in 2022. He is currently pursuing the M.S. degree in electronic and communication engineering with Guilin University of Electronic Technology, China. His current research interests include machine learning and intelligent image processing.
Ronghua Zou received the B.S. degree from Guangzhou Maritime University, China, in 2022. He is currently pursuing the M.S. degree in electronic and communication engineering with Guilin University of Electronic Technology, China. His current research interests include machine learning and intelligent image processing.View more
School of Information and Communication, Guilin University of Electronic Technology, Guilin, China
Hua Yuan received the B.S. degree in computers and applications and the M.S. degree in electronic information engineering from Guilin University of Electronic Technology, China, in 1999 and 2012, respectively. He is currently working as a full-time Lecturer with Guilin University of Electronic Technology. In recent years, he has presided more than three department-level projects. He has authored or co-authored many articl...Show More
Hua Yuan received the B.S. degree in computers and applications and the M.S. degree in electronic information engineering from Guilin University of Electronic Technology, China, in 1999 and 2012, respectively. He is currently working as a full-time Lecturer with Guilin University of Electronic Technology. In recent years, he has presided more than three department-level projects. He has authored or co-authored many articl...View more
School of Information and Communication, Guilin University of Electronic Technology, Guilin, China
Jianwen Mo received the B.S. degree from the Department of Electronic Engineering, North China Electric Power University, in 1994, the M.S. degree from the School of Mathematics and Computer Science, Guangxi Normal University, in 2002, and the Ph.D. degree from the School of Electronic Engineering, Xidian University, in 2011. He is currently working as a full-time Associate Professor with Guilin University of Electronic Technology, China. He presided over and mainly participated in a number of projects, including projects of the National Natural Science Fund and Guangxi Natural Science Fund. His research interests include intelligent information processing, computer vision and image processing, machine learning, and artificial intelligence.
Jianwen Mo received the B.S. degree from the Department of Electronic Engineering, North China Electric Power University, in 1994, the M.S. degree from the School of Mathematics and Computer Science, Guangxi Normal University, in 2002, and the Ph.D. degree from the School of Electronic Engineering, Xidian University, in 2011. He is currently working as a full-time Associate Professor with Guilin University of Electronic Technology, China. He presided over and mainly participated in a number of projects, including projects of the National Natural Science Fund and Guangxi Natural Science Fund. His research interests include intelligent information processing, computer vision and image processing, machine learning, and artificial intelligence.View more
School of Information and Communication, Guilin University of Electronic Technology, Guilin, China
Ronghua Zou received the B.S. degree from Guangzhou Maritime University, China, in 2022. He is currently pursuing the M.S. degree in electronic and communication engineering with Guilin University of Electronic Technology, China. His current research interests include machine learning and intelligent image processing.
Ronghua Zou received the B.S. degree from Guangzhou Maritime University, China, in 2022. He is currently pursuing the M.S. degree in electronic and communication engineering with Guilin University of Electronic Technology, China. His current research interests include machine learning and intelligent image processing.View more
School of Information and Communication, Guilin University of Electronic Technology, Guilin, China
Hua Yuan received the B.S. degree in computers and applications and the M.S. degree in electronic information engineering from Guilin University of Electronic Technology, China, in 1999 and 2012, respectively. He is currently working as a full-time Lecturer with Guilin University of Electronic Technology. In recent years, he has presided more than three department-level projects. He has authored or co-authored many articles in academic journals as the first author and received many national invention patents. His current research interests include intelligent information processing, computer vision and image processing, machine learning, and artificial intelligence.
Hua Yuan received the B.S. degree in computers and applications and the M.S. degree in electronic information engineering from Guilin University of Electronic Technology, China, in 1999 and 2012, respectively. He is currently working as a full-time Lecturer with Guilin University of Electronic Technology. In recent years, he has presided more than three department-level projects. He has authored or co-authored many articles in academic journals as the first author and received many national invention patents. His current research interests include intelligent information processing, computer vision and image processing, machine learning, and artificial intelligence.View more