The research paper presents ACT2G, an attention-based contrastive learning method for text-to-gesture generation. The method uses attention weight for each word in the input text to generate gestures that reflect the content of the text. The study suggests that this approach produces more realistic and human-like gestures, which is important for smooth and natural communication between humans and AI systems. It was found that a wide range of gestures could be generated from the same text by adjusting the attention weights.
Publication date: 28 Sep 2023
Project Page: https://arxiv.org/abs/2309.16162v1
Paper: https://arxiv.org/pdf/2309.16162
