Passive NLOS imaging based on menifold embedding

Input shadow, output reconstruction result and real hidden scene

Passive NLOS imaging is an extremely pathological problem. This project aims to complete data-driven passive NLOS imaging through deep learning. The proposed algorithm can make use of previously underutilized scene priors, thereby improving the effect of passive NLOS imaging.

We also collected a data set with more than 3,000,000 samples, hoping to alleviate the problem of insufficient data set faced by NLOS imaging. After all, the performance of the supervised algorithm depends to a large extent on the quality of the dataset.

This project has been submitted to a journal.

Ruixu Geng (耿瑞旭)
Ruixu Geng (耿瑞旭)
Graduate Student

My research interests include computer vision and computational imaging, especially Non-line-of-sight (NLOS) imaging.