论文标题
上流：用于无监督的光流学习的提升金字塔
UPFlow: Upsampling Pyramid for Unsupervised Optical Flow Learning
论文作者
论文摘要
我们通过改善金字塔网络的上采样和学习来提出一种无监督的学习方法，以进行光流估计。我们设计了一个自引导的Uplample模块，以解决由金字塔水平之间的双线性上采样引起的插值模糊问题。此外，我们提出了金字塔蒸馏损失，以通过将最优质的流量作为伪标签来增加中间水平的监督。通过将这两个组件集成在一起，我们的方法可以在多个领先的基准上（包括MPI-Sintel，Kitti 2012和Kitti 2012和Kitti 2015）实现最佳性能。尤其是，我们在2012年的Kitti上实现了EPE = 1.4，而F1 = 9.38％在Kitti 2015上，在以前的州立大学（Kitti）2015年，以前的州立大学（22.2％）和22％的方法。
We present an unsupervised learning approach for optical flow estimation by improving the upsampling and learning of pyramid network. We design a self-guided upsample module to tackle the interpolation blur problem caused by bilinear upsampling between pyramid levels. Moreover, we propose a pyramid distillation loss to add supervision for intermediate levels via distilling the finest flow as pseudo labels. By integrating these two components together, our method achieves the best performance for unsupervised optical flow learning on multiple leading benchmarks, including MPI-SIntel, KITTI 2012 and KITTI 2015. In particular, we achieve EPE=1.4 on KITTI 2012 and F1=9.38% on KITTI 2015, which outperform the previous state-of-the-art methods by 22.2% and 15.7%, respectively.
