(КПІ ім. Ігоря Сікорського, 2022) Kolodiazhna, O. O.; Uss, M. L.
Stereo matching is one of the core technologies in computer vision, which aims to recover 3D structures of real world from
2D images. Deep stereo networks are one of the best methods to achieve good correspondences between images in stereo
pairs. While there are many available datasets for monocular depth estimation, amount of stereo datasets for training
supervised deep neural networks is limited. In this work, we analyze two different methods to obtain stereo images from a
single RGB image: Neural Radiance Fields (NeRF) and Stereo from Mono algorithm. The results show that the first
method can lead to more accurate solution than the algorithm based on pixels manipulations.