本文介绍FlowNet2.0的改进之处，包括关注训练数据的重要性、开发包含图像形变的堆叠架构以及引入专门处理小位移的子网络。相较于前代，FlowNet2.0仅略微减慢了速度，但将误差降低了超过50%。
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%.
扫一扫
1178
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
