CountLLM introduces a novel approach using large language models to count repetitive actions in videos, addressing limitations of traditional regression networks.

• Enhanced generalization across diverse counting scenarios
• Improved accuracy for variable periodic patterns
• Reduced reliance on narrow training datasets
• Broader applications in health tracking and physical therapy

For healthcare professionals, this technology enables more precise fitness monitoring, rehabilitation assessment, and exercise adherence tracking without specialized equipment—creating opportunities for remote care solutions and personalized health interventions.

CountLLM: Towards Generalizable Repetitive Action Counting via Large Language Model