2019/03/18 by Alex Wong, Wong, Alex, Byung‐Woo Hong +3 · 2 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · Engineering · #Advanced Fluorescence Microscopy Techniques #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Processing Techniques and Applications
paper · pdf · doi:10.48550/arxiv.1903.07309
openalex publication_date 2019/03/18 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Supervised learning methods to infer (hypothesize) depth of a scene from a\nsingle image require costly per-pixel ground-truth. We follow a geometric\napproach that exploits abundant stereo imagery to learn a model to hypothesize\nscene structure without direct supervision. Although we train a network with\nstereo pairs, we only require a single image at test time to hypothesize\ndisparity or depth. We propose a novel objective function that exploits the\nbilateral cyclic relationship between the left and right disparities and we\nintroduce an adaptive regularization scheme that allows the network to handle\nboth the co-visible and occluded regions in a stereo pair. This process\nultimately produces a model to generate hypotheses for the 3-dimensional\nstructure of the scene as viewed in a single image. When used to generate a\nsingle (most probable) estimate of depth, our method outperforms\nstate-of-the-art unsupervised monocular depth prediction methods on the KITTI\nbenchmarks. We show that our method generalizes well by applying our models\ntrained on KITTI to the Make3d dataset.\n