Recently, Professor Wei Ying's team from the College of Information Science and Engineering has made important progress in the research of human-object interaction detection. The research FGAHOI: Fine-Grained Anchors for Human-Object Interaction Detection was published in IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), a top academic journal in the field of artificial intelligence. The first author of the paper is Dr. Ma Shuailei, and the corresponding author is Professor Wei Ying.
Human-object Interaction (HOI) detection is an important issue in the field of computer vision, which needs to identify the interaction between human and objects. Compared with individual object instances, HOI instances span more in space, scale, and tasks, so their detection is more susceptible to noise backgrounds. To alleviate the interference of noise backgrounds on HOI detection, this paper proposes a novel end-to-end framework called FGAHOI, which uses input image information to generate fine-grained anchors to guide HOI detection. The research has improved the method of extracting key features from images with complex background information and integrating the extracted features with query embedding to achieve semantic unity. To promote the development of artificial intelligence, the algorithm code and data of the paper have been set as open source.
In this paper, a novel transformer-based HOI detector called FGAHOI is proposed, which uses input features to generate fine-grained anchors to mitigate the impact of noise backgrounds on the detection of HOI instances. It proposes a novel training strategy, that is, to train each component of the model in order to clarify the training direction of each stage, so as to minimize the training cost. Two new indicators and a new data set, i.e. HOI-SDC, are put forward to address the two challenges of detecting HOI instances, namely the non-uniform distribution region of human-object pairs and the remote vision modeling of human-object pairs. The project team has carried out a lot of experiments on the three benchmark data sets of HHIC-DET, HOI-SDC and V-COCO, and proved the effectiveness of the proposed FGAHOI.
Professor Wei Ying has been committed to image processing and computer vision, medical image computing and analysis, video image analysis and understanding, machine learning and deep learning for a long time. As the project leader, she has directed/completed more than 20 projects including NSFC projects, provincial and ministerial research projects, and enterprise-institution cooperation research projects, and published more than 70 papers in important academic journals/at international conferences at home and abroad. Additionally, she has won more than 20 honors, such as the first/second prize of “Liaoning Province Natural Academic Achievements”, the "JANGHO Teaching Award" of NEU, the “Excellent Instructors in NEU Student Science & Technology Competitions”, and the "Good Supervisors in the Eyes of Postgraduates" of NEU. Professor Wei Ying is also the Vice-chairman of Liaoning Provincial Association for Artificial Intelligence and a standing member of the Special Committee of Smart Medical Care under Chinese Association for Artificial Intelligence.
It is reported that IEEE TPAMI ranks first among the four Class A journals in the field of artificial intelligence identified by the Chinese Computer Federation, and is the top journal in the field of computer vision and pattern recognition, with a five-year average impact factor of 26.7. According to the current popular Google Scholar Citation statistics, IEEE TPAMI ranks first on the list of all computer engineering, electronic engineering and artificial intelligence-related journals with an h5-index of 165 points. It mainly includes original scientific research achievements in artificial intelligence, pattern recognition, computer vision and machine learning. IEEE TPAMI is extremely strict in selecting academic papers, with only about 200 articles accepted every year. IEEE TPAMI is one of the most important academic journals in the field of artificial intelligence, pattern recognition, computer vision and machine learning, and is also the most influential and highest-level journal in the field of information.
