1 College of Intelligence and Computing, Tianjin University
2 Medical School, Nanjing University
†Equal contribution * Corresponding author
Fig 1. We contribute a framework for estimating human poses and shapes from lensless measurements. The first three columns show the experimental results on various datasets (captured by our lensless imaging system), and the last two columns show the experimental results on real scenes.
We present a novel approach to human pose and shape estimation through a lensless imaging system. Existing methods for human pose and shape estimation are done with RGB images captured by conventional cameras. While, lensless imaging system can have a smaller size, simpler structure and stronger privacy protection attributes, and hence can be adapted to a variety of complex environments. In this paper, we propose the first end-to-end framework to recover 3D human poses and shapes from lensless measurements, to our best knowledge. We specifically design a multi-level lensless feature decoder to decode the lensless measurements base on mask-encoder to obtain more efficient features. We also propose a dual auxiliary supervision mechanism to improve the accuracy of estimation of human limb ends. Besides, we establish a lenless imaging system and verify the effectiveness of our methed on various datasets acquired by our lensless imaging system. The code and dataset will be will be available for research purposes.
Fig 2. Overview of the proposed framework. A measurement M is passsed through a Multi-Scale Lensless Feature Decoder to get spatial characteristics at different scales. These feature maps are fed into the regressor for human pose and shape estimation. Also these feature maps are fed into the Double-Head Auxiliary Supervision, in order to improve the accuracy of the estimation.
Haoyang Ge, Qiao Feng, Hailong Jia, Xiongzheng Li, Xiangjun Yin, You Zhou, Jingyu Yang, Kun Li. "LPSNet: End-to-End Human Pose and Shape Estimation with Lensless Imaging". In Proc. CVPR, 2024.
@article{ge2024LPSNet,
author = {Haoyang Ge, Qiao Feng, Hailong Jia, Xiongzheng Li, Xiangjun Yin, You Zhou, Jingyu Yang, Kun Li},
title = {LPSNet: End-to-End Human Pose and Shape Estimation with Lensless Imaging},
booktitle = {CVPR},
year={2024},
}