2018/07/28 by Andrea Pilzer, Pilzer, Andrea, Dan Xu +7 · 2 citations
Computer Science · Engineering · #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Processing Techniques and Applications #Optical measurement and interference techniques
paper · pdf · doi:10.48550/arxiv.1807.10915
openalex publication_date 2018/07/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
While recent deep monocular depth estimation approaches based on supervised\nregression have achieved remarkable performance, costly ground truth\nannotations are required during training. To cope with this issue, in this\npaper we present a novel unsupervised deep learning approach for predicting\ndepth maps and show that the depth estimation task can be effectively tackled\nwithin an adversarial learning framework. Specifically, we propose a deep\ngenerative network that learns to predict the correspondence field i.e. the\ndisparity map between two image views in a calibrated stereo camera setting.\nThe proposed architecture consists of two generative sub-networks jointly\ntrained with adversarial learning for reconstructing the disparity map and\norganized in a cycle such as to provide mutual constraints and supervision to\neach other. Extensive experiments on the publicly available datasets KITTI and\nCityscapes demonstrate the effectiveness of the proposed model and competitive\nresults with state of the art methods. The code and trained model are available\non https://github.com/andrea-pilzer/unsup-stereo-depthGAN.\n