ECCV2022|StyleSwap - 基于StyleGAN2的高保真换脸!论文速递2022.9.20!

2022-12-11 12:09:14 浏览数 (1)

整理:AI算法与图像处理

CVPR2022论文和代码整理:https://github.com/DWCTOD/CVPR2022-Papers-with-Code-Demo

ECCV2022论文和代码整理:https://github.com/DWCTOD/ECCV2022-Papers-with-Code-Demo

最新成果demo展示:

ECCV2022|StyleSwap - 基于StyleGAN2的高保真换脸!

主页: https://hangz-nju-cuhk.github.io/projects/StyleSwap

摘要:

鉴于其广泛的应用,已经对不可知的人脸交换任务进行了许多尝试。虽然现有的方法大多依赖于繁琐的网络和损失设计,但它们仍然在源和目标人脸之间的信息平衡方面存在困难,并且往往会产生可见的伪影。在这项工作中,我们介绍了一个简洁有效的框架 StyleSwap。我们的核心思想是利用基于样式的生成器来实现高保真和鲁棒的人脸交换,从而可以利用生成器的优势来优化身份相似度。我们发现,只需进行最少的修改,StyleGAN2 架构就可以成功处理来自源和目标的所需信息。此外,受 ToRGB 层的启发,进一步设计了交换驱动掩码分支以改进信息混合。此外,可以采用 StyleGAN 反转的优势。特别是,提出了一种交换引导的 ID 反转策略来优化身份相似性。大量实验验证了我们的框架产生了高质量的人脸交换结果,在质量和数量上都优于最先进的方法。


最新论文整理

ECCV2022

Updated on : 20 Sep 2022
total number : 11

Real-time Online Video Detection with Temporal Smoothing Transformers

  • 论文/Paper: http://arxiv.org/pdf/2209.09236
  • 代码/Code: https://github.com/zhaoyue-zephyrus/TeSTra

3D-PL: Domain Adaptive Depth Estimation with 3D-aware Pseudo-Labeling

  • 论文/Paper: http://arxiv.org/pdf/2209.09231
  • 代码/Code: None

Panoramic Vision Transformer for Saliency Detection in 360° Videos

  • 论文/Paper: http://arxiv.org/pdf/2209.08956
  • 代码/Code: None

HVC-Net: Unifying Homography, Visibility, and Confidence Learning for Planar Object Tracking

  • 论文/Paper: http://arxiv.org/pdf/2209.08924
  • 代码/Code: None

D&D: Learning Human Dynamics from Dynamic Camera

  • 论文/Paper: http://arxiv.org/pdf/2209.08790
  • 代码/Code: https://github.com/Jeffsjtu/DnD

RVSL: Robust Vehicle Similarity Learning in Real Hazy Scenes Based on Semi-supervised Learning

  • 论文/Paper: http://arxiv.org/pdf/2209.08630
  • 代码/Code: None

ActiveNeRF: Learning where to See with Uncertainty Estimation

  • 论文/Paper: http://arxiv.org/pdf/2209.08546
  • 代码/Code: url{https://github.com/LeapLabTHU/ActiveNeRF}.

MIPI 2022 Challenge on RGBW Sensor Re-mosaic: Dataset and Report

  • 论文/Paper: http://arxiv.org/pdf/2209.08471
  • 代码/Code: https://github.com/mipi-challenge/MIPI2022.

Learning to Weight Samples for Dynamic Early-exiting Networks

  • 论文/Paper: http://arxiv.org/pdf/2209.08310
  • 代码/Code: https://github.com/LeapLabTHU/L2W-DEN.

Few-Shot Classification with Contrastive Learning

  • 论文/Paper: http://arxiv.org/pdf/2209.08224
  • 代码/Code: None

PPT: token-Pruned Pose Transformer for monocular and multi-view human pose estimation

  • 论文/Paper: http://arxiv.org/pdf/2209.08194
  • 代码/Code: https://github.com/HowieMa/PPT
CVPR2022

0 人点赞