0%

about me

Resume

Hi, 这里是 迪三 (Disanda) 的 个人介绍。

News

关注我的微信公众号 <<迪AI>>, 聚焦AIGC新范式。

Yu Cheng

Education

  • Chongqing University of Posts and Telecommunications (2007.09 - 2011.06)

Bechelor | Computer Science and Technology | Chongqing

  • Northeastern University (2014.09 ~ 2017.01)

Master | Computer Application and Technology | Shenyang

  • Macau University of Science and Technology (2019.09 ~ 2022.09)

PhD | Computer Technology and Application | Macao

Publication

以下是博士期间的研究:

  • “Fast Transformation of Discriminators into Encoders using Pre-Trained GANs”, C. Yu and W. Wang, Pattern Recognition Letters, 2021, 0167-8655, JCR Q2, IF:3.756, [Paper] | [Code: AnimationGAN].

这项工作灵感来自InfoGAN, 更深入的探讨如何用自监督的方式让数据集(shuffle或者无序的)呈现动画效果。

  • “Self-Supervised Animation Synthesis Through Adversarial Training”, C. Yu, W. Wang and J. Yan, IEEE Access, vol. 8, pp. 128140-128151, 2020, JCR Q2, IF:3.367, [Paper] | [Code: Sym:GAN Encoder].

这项工作应用深度网络的预训练模型(Pre-trained GAN), 将判别器“改造“为生成器对应的编码器,以重构图片。

  • “Adaptable GAN Encoders for Image Reconstruction via Multi-type Latent Vectors with Two-scale Attentions”, C. Yu and W. Wang, arXiv eprint, 2108.10201, [Paper] | [Code: MTA-TSA].

这项工作做一种适配型的编码器,用于重构当下主流deep GANs的合成图片(e.g., PGGAN, StyleGANs, BigGAN), 并探索将真实高清人脸图片映射到stylegGANs中, 为后续做人脸图像编辑做准备。

  • “2-Step Regularization on Style Optimization for Real Face Morphing”, Cheng Yu; Wenmin Wang; HongLei Li; Bugiolacchi, Roberto, IEEE TechRxiv, 2022. [Paper]

上一份工作的后续。通过探索潜空间向量和真实人脸图片的映射关系,实现真实人脸的表情编辑。相比baseline,本工作在人脸表情编辑上表现更为多样丰富,属性表征更为细腻逼真。实现原理基于两个人脸属性分类器。这里归纳,整理并重构了当前人脸分类属性集中的数据结构。最终实现来自逻辑回归(潜向量和属性对的分类)。

  • “SwapInpaint: Identity-specific Face Inpainting with Identity Swapping”, IEEE Transactions on Circuits and Systems for Video Technology, H. Li, W. Wang, C. Yu and S. Zhang. JCR Q1, IF:4.685, [Paper].

这项工作主要研究了人脸图像修复,并提出了一种对当下baseline改进的方法。本人部分参与了论文前期的Discussion和Experiment. 以及投稿期间的Revision和 Rebuttal等。