It's model from a A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimation article. I used it to train DispNet model for solving a stereo disparity problem. Using a FlyingThings3D dataset I obtained endpoint error about 2.8. During training I used next random augmentations: crop, rotation in range -5 to 5 degrees, color jitter, gaussian noise. I did not make any hyperparamters search because of lack of computing resources and time.
Having two images of a scene from left and right cameras we want to get depth map of the scene.