Lane Detection in Low-light Conditions Using an Efficient Data Enhancement : Light Conditions Style Transfer

Nowadays, deep learning techniques are widely used for lane detection, but\napplication in low-light conditions remains a challenge until this day.\nAlthough multi-task learning and contextual-information-based methods have been\nproposed to solve the problem, they either require additional manual\nannotations or introduce extra inference overhead respectively. In this paper,\nwe propose a style-transfer-based data enhancement method, which uses\nGenerative Adversarial Networks (GANs) to generate images in low-light\nconditions, that increases the environmental adaptability of the lane detector.\nOur solution consists of three parts: the proposed SIM-CycleGAN, light\nconditions style transfer and lane detection network. It does not require\nadditional manual annotations nor extra inference overhead. We validated our\nmethods on the lane detection benchmark CULane using ERFNet. Empirically, lane\ndetection model trained using our method demonstrated adaptability in low-light\nconditions and robustness in complex scenarios. Our code for this paper will be\npublicly available.\n