- 博客(4)
- 收藏
- 关注
原创 AOT
处理appearance gap很大的图像伪造,比如说复杂的光照条件和肤色 把appearance mapping 定义为一个最优传输问题,在特征空间和像素空间 a relighting generator来模拟最优传输,1通过求特征空间中学习到的特征的Wasserstein distance最小值。2在像素空间采用一种segmentation game to 最小化Wasserstein distance。 通过操纵直方图分布,即将源图...
2021-03-31 09:03:13 197
原创 FaceShifter: Towards High Fidelity And Occlusion Aware Face Swapping
FaceShifter: Towards High Fidelity And Occlusion Aware Face Swapping摘要两阶段框架,在第一阶段就彻底的和自适应的提取和集成target属性(head pose,facial expression, lighting, background )。我们提出了一个提取 multi-level target face attribute 的一个attribute encoder。一个Adaptive Attention Denorma
2021-02-21 20:30:03 356
原创 Everybody Dance Now
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 deta
2020-12-21 21:39:53 215
原创 ConvTransformer: A Convolutional Transformer Network for Video Frame
论文摘抄好词好句好的句子好的句子1.Deep Convolutional Neural Networks (CNNs) are powerful models that have achieved excellent performance on difficult computer vision tasks.2.Although CNNS perform well whenever large labeled training samples are available,they work badl
2020-12-04 20:17:42 1604
空空如也
空空如也
TA创建的收藏夹 TA关注的收藏夹
TA关注的人