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 the estimation error by more than 50%.