mini-Unet GAN: Optimized GAN for Monocular Depth Estimation

Depth estimation is an ill-posed problem in computer vision which has numerous applications such as augmented reality, virtual reality, and autonomous driving. Recently, with the emergence of deep learning, depth estimation has achieved significant results. However, deep learning-based methods for depth estimation require heavy computational resources. This challenge limits the scalability of deploying depth estimation models in applications such as autonomous driving where efficient, and accurate depth maps are crucial for safety. To tackle this problem, in this paper, we propose a novel efficient network based on generative adversarial networks named mini-Unet GAN to recover depth from monocular images. First, we propose a novel strategy to optimize a famous generative adversarial network to reduce its complexity. Then, we propose a set of techniques to increase the accuracy of the model to generate precise depth maps. Extensive experiments were conducted on KITTI dataset and real-life images and comparative evaluations with the state of the art are presented.