arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1910.04149cs.CV
收藏

具有循环一致性的非对齐图像到序列的变换

Unaligned Image-to-Sequence Transformation with Loop Consistency

University of California San Diego(加州大学圣地亚哥分校)

Siyang Wang, Justin Lazarow, Kwonjoon Lee, Zhuowen Tu

摘要:

我们处理建模顺序视觉现象的问题。给定可以划分为离散时间步的现象的例子,我们的目标是从任何这样的时间获取输入,并在序列中的所有其他时间步实现该输入。此外,我们的目标是在没有地面实况比对序列的情况下做到这一点-避免收集比对数据所需的困难。这将不成对的图像到图像问题从生成对推广到生成序列。我们扩展循环的一致性循环的一致性,并减轻学习的困难,在由此产生的长链的计算。我们显示竞争力的结果相比,现有的图像到图像的技术建模时,几个不同的数据集,包括地球的季节和老化的人脸。

英文摘要:

We tackle the problem of modeling sequential visual phenomena. Given examples of a phenomena that can be divided into discrete time steps, we aim to take an input from any such time and realize this input at all other time steps in the sequence. Furthermore, we aim to do this without ground-truth aligned sequences -- avoiding the difficulties needed for gathering aligned data. This generalizes the unpaired image-to-image problem from generating pairs to generating sequences. We extend cycle consistency to loop consistency and alleviate difficulties associated with learning in the resulting long chains of computation. We show competitive results compared to existing image-to-image techniques when modeling several different data sets including the Earth's seasons and aging of human faces.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑