New framework KineMIC enhances few-shot action synthesis for HAR

Researchers have developed KineMIC, a novel transfer learning framework designed to improve few-shot action synthesis for skeletal-based Human Activity Recognition (HAR). This method adapts text-to-motion diffusion models, which are typically trained for general artistic motion, to the specific needs of HAR by leveraging CLIP text embeddings for kinematic distillation. By using only 10 samples per action class from the NTU RGB+D 120 dataset, KineMIC significantly enhances motion coherence and provides a robust data augmentation source, leading to a 23.1% improvement in accuracy for HAR classifiers. AI

IMPACT This research could lead to more efficient and effective training of AI models for understanding human activities from limited data.

RANK_REASON The cluster contains a research paper detailing a new framework for action synthesis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Human Activity Recognition
• HumanML3D
• Luca Cazzola
• text-to-motion diffusion models

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →