The ACM International Multimedia Conference (ACM Multimedia 2023) was held in Ottawa, Canada, on October 29, 2023. The ACM International Multimedia Conference is the only A-class international top conference in the multimedia research field certified by the China Computer Federation (CCF). The conference held a series of challenge competitions focusing on research hotspots in the multimedia field such as affective computing and group behavior analysis. The teachers and students from the School of Computer and Information at Hefei University of Technology participated in the competition and won four championships in the Multimodal Emotion Analysis Challenge and the Multimodal Group Behavior Analysis Challenge. The specific situations of different tracks are as follows:

(1) ACMMM2023 Multimodal Emotion Analysis Challenge:

The International Multimodal Emotion Analysis Challenge is the fourth challenge hosted by the ACM Multimedia Conference, co-sponsored by Imperial College London, University of Augsburg in Germany, and Nanyang Technological University, among others. It is reported that in 2023, 31 teams from 25 research institutions in 8 countries around the world participated, including top teams in the field from the Institute of Automation of the Chinese Academy of Sciences, University of New South Wales, Ghent University, University of Chinese Academy of Sciences, SenseTime Research Institute, iFLYTEK, and Ximalaya. This year's Multimodal Emotion Analysis Challenge is divided into three tracks: Humor Detection, Imitation Emotion Recognition, and Personalized Emotion Analysis. The competition focuses on multimodal high-robust compound emotion analysis in natural scenes, emphasizing the evaluation of multimodal deep fusion capabilities combined with semantic information. Participants need to effectively predict psychological indicators such as humor, emotion analysis, and emotional reactions based on different datasets provided by the official for different challenge content, including audio and video data in natural scenes, text data, and physiological signal data.

The School of Computer and Information, led by Professor Wang Meng and under the guidance of teachers such as Sun Xiao, formed a team of several master's and doctoral students to participate in the competition. They won the championship in the Personalized Emotion Analysis track, discovered two efficient visual features for the problem of personalized emotion analysis, and proposed an effective multimodal fusion solution, achieving excellent performance in complex personalized emotion detection and winning first place. At the same time, they proposed an attention-based multimodal feature fusion model in the Humor Detection Challenge and also won first place in the Humor Detection track. The team published papers at the conference and introduced their work online.

(2) ACMMM2023 Multi-modal Group Behavior Analysis Challenge:

The Multi-modal Group Behavior Analysis Challenge (MultiMediate: Multi-modal Group Behaviour Analysis for Artificial Mediation) is one of the grand challenges (only 10 per conference) of the ACM Multimedia 2023 conference, co-sponsored by the German Research Center for Artificial Intelligence, University of Stuttgart, University of Augsburg, and the French National Institute for Computer and Automation Research, among others. Reportedly, in 2023, it attracted top teams in the field from more than a dozen research institutions around the world, including Mohammed Bin Zayed University of Artificial Intelligence, University of Canberra, Monash University, Hong Kong Polytechnic University, University of Science and Technology of China, Hunan University, Suzhou University, and Yunzhisheng Technology Co., Ltd. This year's Multi-modal Group Behavior Analysis Challenge is divided into six tracks: Body Behavior Recognition, Engagement Estimation, Backchannel Detection, Consistency Estimation, Eye Contact Detection, and Next Speaker Prediction. The competition focuses on group behavior perception and analysis tasks, with this year's challenge focusing on body behavior recognition and engagement estimation. Participants need to use different datasets provided by the official according to different challenge content to effectively detect group behavior.

The team from the Multimedia Computing Laboratory of the School of Computer and Information at Hefei University of Technology, led by Professor Wang Meng and guided by Professor Guo Dan, consisted of several master's and doctoral students including Li Kun, Chen Guoliang, and Liu Feiyang. They won the championship in both the Body Behavior Recognition and Eye Contact Detection tracks. They proposed an effective data augmentation learning scheme for action recognition problems, achieving excellent performance in complex multi-category body behavior recognition and eye contact detection, and won first place in both tracks. The team published papers at the conference and introduced their work online.