Lijun Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Lijun Zhang has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Si Liu
Beihang University
2 shared publications - Chen Fu
RoomF217, new main building, Beihang University, No.37, xueyuanroad, HaiDian District, BeiJing, China
1 shared publication - Haihang Xu
Department of Computer Science Beijing University of Aeronautics and Astronautics 100191 Beijing China
1 shared publication - Luting Wang
Beihang University
1 shared publication - Wenjun Wu
School of Artificial Intelligence, Beihang University, Beijing 100191, China
1 shared publication - Yan Wang
Tsinghua University
1 shared publication
Latest Publications
- 25 October 2024
- Published by Springer Nature
- p. 22-38
Asynchronous Large Language Model Enhanced Planner for Autonomous Driving
- 25 October 2024
- Published by Springer Nature
- p. 22-38
Asynchronous Large Language Model Enhanced Planner for Autonomous Driving
- Open Access
Knowledge Distillation via Query Selection for Detection Transformer
Transformers have revolutionized the object detection landscape by introducing DETRs, acclaimed for their simplicity and efficacy. Despite their advantages, the substantial size of these models poses significant challenges for practical deployment, particularly in resource-constrained environments. This paper addresses the challenge of compressing DETR by leveraging knowledge distillation, a technique that holds promise for maintaining model performance while reducing size. A critical aspect of DETRs' performance is their reliance on queries to interpret object representations accurately. Traditional distillation methods often focus exclusively on positive queries, identified through bipartite matching, neglecting the rich information present in hard-negative queries. Our visual analysis indicates that hard-negative queries, focusing on foreground elements, are crucial for enhancing distillation outcomes. To this end, we introduce a novel Group Query Selection strategy, which diverges from traditional query selection in DETR distillation by segmenting queries based on their Generalized Intersection over Union (GIoU) with ground truth objects, thereby uncovering valuable hard-negative queries for distillation. Furthermore, we present the Knowledge Distillation via Query Selection for DETR (QSKD) framework, which incorporates Attention-Guided Feature Distillation (AGFD) and Local Alignment Prediction Distillation (LAPD). These components optimize the distillation process by focusing on the most informative aspects of the teacher model's intermediate features and output. Our comprehensive experimental evaluation of the MS-COCO dataset demonstrates the effectiveness of our approach, significantly improving average precision (AP) across various DETR architectures without incurring substantial computational costs. Specifically, the AP of Conditional DETR ResNet-18 increased from 35.8 to 39.9.
Open AccessKnowledge Distillation via Query Selection for Detection Transformer
Transformers have revolutionized the object detection landscape by introducing DETRs, acclaimed for their simplicity and efficacy. Despite their advantages, the substantial size of these models poses significant challenges for practical deployment, particularly in resource-constrained environments. This paper addresses the challenge of compressing DETR by leveraging knowledge distillation, a technique that holds promise for maintaining model performance while reducing size. A critical aspect of DETRs' performance is their reliance on queries to interpret object representations accurately. Traditional distillation methods often focus exclusively on positive queries, identified through bipartite matching, neglecting the rich information present in hard-negative queries. Our visual analysis indicates that hard-negative queries, focusing on foreground elements, are crucial for enhancing distillation outcomes. To this end, we introduce a novel Group Query Selection strategy, which diverges from traditional query selection in DETR distillation by segmenting queries based on their Generalized Intersection over Union (GIoU) with ground truth objects, thereby uncovering valuable hard-negative queries for distillation. Furthermore, we present the Knowledge Distillation via Query Selection for DETR (QSKD) framework, which incorporates Attention-Guided Feature Distillation (AGFD) and Local Alignment Prediction Distillation (LAPD). These components optimize the distillation process by focusing on the most informative aspects of the teacher model's intermediate features and output. Our comprehensive experimental evaluation of the MS-COCO dataset demonstrates the effectiveness of our approach, significantly improving average precision (AP) across various DETR architectures without incurring substantial computational costs. Specifically, the AP of Conditional DETR ResNet-18 increased from 35.8 to 39.9.
- Open Access
- 31 July 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (15) , 7100
Online Blended Learning in Small Private Online Course
In this work, we studied the online blended learning model of computer network experimentation, focusing mainly on the problem of traditional network experiments being limited by location and time, and explore the applicability of the small private online course (SPOC) advanced teaching concepts to computer network online experiment teaching. Based on the structure of a combination of virtual and real, real and not virtual, an online network experiment platform and management system has been designed and constructed, enabling students to carry out remote online computer network hardware experiments anytime and anywhere, without being restricted by time, space, or content. Using the online network experiment platform, we can organize the experimental modules and knowledge points via the SPOC course concept, by developing online network experimental content, modularizing and fragmenting of the experiments, creating the pre-experimental explanation and experiment preview videos, and evaluating the assignments via peer grading to analyze students’ learning behavior. By exploring online network experimental teaching methods and management models, offering experimental guidance in an interactive manner, and highlighting the openness and sharing characteristics of online experimental teaching platforms, we can improve the utilization rate for teaching resources, and provide ideas for applied scientific research methods.
- 31 July 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (15) , 7100
Open AccessOnline Blended Learning in Small Private Online Course
In this work, we studied the online blended learning model of computer network experimentation, focusing mainly on the problem of traditional network experiments being limited by location and time, and explore the applicability of the small private online course (SPOC) advanced teaching concepts to computer network online experiment teaching. Based on the structure of a combination of virtual and real, real and not virtual, an online network experiment platform and management system has been designed and constructed, enabling students to carry out remote online computer network hardware experiments anytime and anywhere, without being restricted by time, space, or content. Using the online network experiment platform, we can organize the experimental modules and knowledge points via the SPOC course concept, by developing online network experimental content, modularizing and fragmenting of the experiments, creating the pre-experimental explanation and experiment preview videos, and evaluating the assignments via peer grading to analyze students’ learning behavior. By exploring online network experimental teaching methods and management models, offering experimental guidance in an interactive manner, and highlighting the openness and sharing characteristics of online experimental teaching platforms, we can improve the utilization rate for teaching resources, and provide ideas for applied scientific research methods.
- Open Access
- 5 September 2017
- Published by EDP Sciences in ITM Web of Conferences
- Vol. 12, 03028
Adaptive Load Balancing Strategy Based on LVS
ITM Web of Conferences, open-access proceedings in information technology, computer science and mathematics
- 5 September 2017
- Published by EDP Sciences in ITM Web of Conferences
- Vol. 12, 03028
Open AccessAdaptive Load Balancing Strategy Based on LVS
ITM Web of Conferences, open-access proceedings in information technology, computer science and mathematics
- Open Access
- 1 January 2012
- Published by Atlantis Press SARL
Application of Link Prediction in Temporal Networks
- 1 January 2012
- Published by Atlantis Press SARL
Open AccessApplication of Link Prediction in Temporal Networks
- Open Access
- 1 January 2012
- Published by Atlantis Press SARL
Design and Implementation of Positioning and Navigation System Based on Android
- 1 January 2012
- Published by Atlantis Press SARL
Open AccessDesign and Implementation of Positioning and Navigation System Based on Android