Fast Non-Line-Of-Sight Imaging with Two-Step Deep Remapping

Conventional imaging only records photons directly sent from the object to detector, while non-line-of-sight (NLOS) takes indirect light into account. Most NLOS solutions employ a transient scanning process, followed by physical based algorithm reconstruct scenes. However, detection requires sophisticated apparatus, with long time and low robustness ambient environment, reconstruction algorithms are typically time-consuming computationally expensive. Here we propose new solution address above defects, innovations on both equipment algorithm. We apply inexpensive commercial Lidar for detection, much higher speed better compatibility real-world imaging. Our framework is deep learning based, generative two-step remapping strategy guarantee high fidelity. The overall process allows millisecond responses, precision of millimeter level. have experimentally tested proposed synthetic real objects, further demonstrated our method be applicable full-color