• Author(s): Zeren Jiang, Chen Guo, Manuel Kaufmann, Tianjian Jiang, Julien Valentin, Otmar Hilliges, Jie Song

This paper presents MultiPly, a novel approach for reconstructing multiple people from monocular video captured in uncontrolled environments. Traditional methods for 3D human reconstruction often rely on multi-view setups or controlled conditions, limiting their applicability in real-world scenarios. MultiPly addresses these limitations by enabling accurate 3D reconstruction of multiple individuals from a single camera view in diverse and dynamic settings.

The proposed method leverages a combination of advanced computer vision techniques and deep learning models to achieve high-quality reconstructions. MultiPly employs a multi-stage pipeline that includes person detection, pose estimation, and 3D reconstruction. Initially, the system detects and tracks individuals in the video using a robust person detection algorithm. Subsequently, a state-of-the-art pose estimation model predicts the 2D joint locations for each detected person. These 2D poses are then used to infer the 3D structure through a novel optimization framework that ensures consistency and accuracy across frames.

Extensive experiments are conducted to evaluate the performance of MultiPly on various datasets, including challenging real-world videos with multiple people and complex interactions. The results demonstrate that MultiPly significantly outperforms existing methods in terms of reconstruction accuracy and robustness. The system is capable of handling occlusions, varying lighting conditions, and diverse poses, making it suitable for a wide range of applications.

The paper also explores potential applications of MultiPly in fields such as sports analytics, virtual reality, and human-computer interaction. By providing a reliable solution for 3D human reconstruction from monocular video, MultiPly opens up new possibilities for analyzing and understanding human activities in natural environments. The authors discuss the implications of their findings and suggest directions for future research to further enhance the capabilities of the proposed method. MultiPly represents a significant advancement in the field of 3D human reconstruction, offering a practical and effective solution for reconstructing multiple people from monocular video in the wild. The proposed approach addresses the limitations of traditional methods and paves the way for new applications in various domains.