解读：FlowNet 2.0 evolution of optical flow estimation with deep networks

贡献点:

1. focus on the training data and show that the schedule of presenting data during training is very important. 

2. develop a stacked architecture that includes warping of the second image with intermediate optical flow. 

3. we elaborate on small displacements by introducing a sub-network specializing on small motions. 

实验结果：

FlowNet 2.0 is only marginally slower than the original FlowNet but decreases theestimation error by more than 50%.