Human Body Shape Estimation
- 0 Collaborators
Estimated shapes can in turn be used for applications such as surveillance, biometric authentication, image retouching, rendering novel viewpoints and also pose estimation, since the integration of body shape knowledge simplifies and improves pose estimation algorithms. A current trend is that of medical and personal measurements. A practical human body shape estimation algorithm should be accurate, robust, efficient and automatic. The existing algorithms do not satisfy these fundamental proper- ties simultaneously. More accurate methods rely on manual input and a fitting pose, while others operate under more restrictive assumptions, or utilize handcrafted features. As a further common shortcoming, most methods have prohibitive time complexity for practical applications. ...learn more
Project status: Concept
Intel Technologies
Intel Opt ML/DL Framework
Overview / Usage
Human body shape estimation is an important problem in computer vision, but has so far not received as much attention as the closely related problems such as pose estimation. The methods so far rely on hand-crafted features and specialized algorithms with possible manual interaction. In contrast, it has been shown repeatedly that utilizing neural networks can lead to superior results for many problems such as classification, segmentation, pose estimation and shape classification or retrieval. However, applying this technique to body shape estimation has not been widely discussed so far. Estimated shapes can in turn be used for applications such as surveil- lance, biometric authentication, image retouching, rendering novel viewpoints and also pose estimation, since the integration of body shape knowledge simplifies and improves pose estimation algorithms. A current trend is that of medical and personal measurements.
Methodology / Approach
Convolutional Neural Networks (CNNs).
Technologies Used
TensorFlow.