Everybody Dance Now
好词
first-of-its-kind 首发的
twist 问题
as is obvious from the figure 从图中可以明显看出
off-the-shelf 现成的
in-the-wild 原始的
Motivated by the high quality of our results 在高质量结果的激励下
Furthermore 此外
a large collection of 许多
For more details, see our project website
in-the-wild 野生的,户外的
The central contribution of our work 我们的重要贡献是
high-resolution 高分辨率
open-source开源
state-of-the-art 最先进的
subtle 细微的(影响)
好句
Abstract
1.Although our method is quite simple, it produces surprisingly compelling results (see video).
Introduction
2.The twist is that the target person never performed the same exact sequence of motions as the source, and, indeed, knows nothing about ballet.
3.He was instead filmed performing a set of standard moves, without specific reference to the precise actions of the source.
4.And, as is obvious from the figure, the source and the target are of different genders, have different builds, and wear different clothing.
- Our goal is, therefore, to discover an image-to-image translation [16] between the source and target sets.
6.To transfer motion from source to target, we input the pose stick figures from the source into the trained model to obtain images of the target subject in the same pose as the source
7.The central contribution of our work is a surprisingly simple method for generating compelling results on human motion transfer
8.We strongly believe that it is important for work in image synthesis to explicitly address the issue of fake detection
9.We observe that keypoint-based pose preserves motion signatures over time while abstracting away as much subject identity as possible and can serve as an intermediate representation between any two subjects.
Related work
1.Over the last two decades there has been extensive work dedicated to motion transfer.
2.Classic computer graphics approaches to motion transfer attempt to perform this in 3D
- Similarly to our method, recent works have applied deep learning for reanimation in different applications and rely on more detailed input representations
4.Rather than generating possible views of a previously unseen person from a single input image, we are interested in learning the style of a single, known person from large amounts of personalized video data and synthesizing them dancing in a detailed high-resolution video.
5.Our work is made possible by recent rapid advances along two separate directions
我们的工作之所以成为可能,是因为最近在两个不同方向上的快速进展
Method
- See Figure 3 for an overview of our pipeline.
- Details of this process are described in the supplementary material.