The 2nd Challenge on Micro-gesture Analysis for Hidden Emotion Understanding (MiGA) 2024: Dataset and Results

This paper summarizes the 2nd Challenge of Micro-gesture Analysis for Hidden Emotion Understanding (MiGA) 2024. The competition was split into two independent tracks: micro-gesture classification from pre-segmented data clips, and micro-gesture online recognition in sequences of continuous data. In this edition of the MiGA challenge, both tracks use multi-modal data (RGB and skeleton as modalities). For evaluation, accuracy for classification and F1 score for online recognition are used as the evaluation measure. Two large micro-gesture datasets (iMiGUE and SMG) were made publicly available and the Kaggle platform was used to manage the competition. Results achieved a classification accuracy of 70.25% for micro-gesture classification, showing a significant improvement compared to last year’s competition, meanwhile, an F1 score for online recognition is about 0.2757 was achieved for multi-modal gesture recognition, showing the task is still challenging and leaves considerable margin for improvement.