基于单幅图像，采用马尔科夫随机场进行三维重建，即推测方向。本文仅假设场景由多个小平面组成，算法将图像分割为若干超像素块，同时考虑深度信息和不同超像素间的关系，进行有监督学习。(rnn_trainnet.m is a scr ipt used for training the S-RNN rnn_testnet.m is a scr ipt used for generating a summary of multiple albums