Two tables: the first showing the accuracy of the system for classification, which is between 44-81% depending on gesture category and model (2.0 vs 2.5, context vs no context). The second table shows that timestamping for gesture onset is very accurate, between 84-100% successful depending on gesture category and model.
Results from my pilot study on Gemini 2.5's ability to automatically classify Co-Speech gesture in one-gesture, one-actor video clips are in. Timestamps ✅, Classification 🟨 44-81% depending on gesture category. #Gemini #CognitiveScience #gesture #cospeechgesture #multimodal #LLM #copilot #AI #ML
3
1
0
0