CAGAN: A Channel-aware Generative Adversarial Network for Passive Non-Line-of-Sight Imaging

This paper proposes a novel network named CAGAN, which is used for passive non-line-of-sight (NLOS) imaging. We used the generative adversarial network (GAN) network for image reconstruction in the passive NLOS field for the first time. We introduce a novel module channel-aware feature fusion module (CAFE) and insert it into the backbone network to capture the interdependence among all channels through their similarity. It adaptively highlights important reconstruction information and suppresses background interference factors. The proposed CAGAN is evaluated on the public data set NLOS-Passive, and the results show that CAGAN is superior to the existing methods in SSIM, PSNR, and CSE.