GAN-Based Day-to-Night Image Style Transfer for Nighttime Vehicle Detection

THE major cause of traffic accidents is mainly due to improper following distance and distracted driving. The most critical function in the advanced driver assistance systems (ADAS) and autonomous vehicles is vehicle detection. One expects that vehicles around host driver could be detected as accurately as possible by an ADAS all day, including day and night. However, vehicle’s appearance at daytime is quite different from its counterpart at nighttime. The existing was AugGAN, a GAN based data augmenter which could transform on-road driving images to a desired domain while image-objects would be well preserved. We quantitatively evaluate different methods by training Faster R-CNN and YOLO with datasets generated from the transformed results and demonstrate significant improvement on the object detection accuracies by using the proposed AugGAN model. In this paper, for images in source domain to be properly translated to the target one while image-objects are well-preserved. We will try to explicitly vector in order to gain multi-modality in performing unpaired image – to – image translation