Towards Real-Time Multi-Person Pose Estimation via Feature Selection and Sharpening Mechanisms

Key Points

• To enhance real-time multi-person pose estimation by utilizing feature selection and sharpening mechanisms.
• Utilized neural network architectures for pose estimation.
• Implemented feature selection to improve detection accuracy.
• Applied sharpening mechanisms to refine joint coordinate regression.
• Evaluated performance in dynamic scenes with multiple human instances.
• Achieved enhanced accuracy in detecting multiple poses simultaneously.
• Demonstrated effective regression of joint coordinates in real-time.
• Improved overall performance metrics compared to existing methods.