A5001407592- Generative Adversarial Networks and Image Synthesis
- Face recognition and analysis
- Advanced Image Processing Techniques
- Chinese history and philosophy
- Asian Culture and Media Studies
- Domain Adaptation and Few-Shot Learning
- Topic Modeling
- Japanese History and Culture
- Advanced Graph Neural Networks
- Digital Media Forensic Detection
- Innovative Educational Techniques
- Image Processing Techniques and Applications
- Advanced Image Fusion Techniques
- 3D Shape Modeling and Analysis
- Video Surveillance and Tracking Methods
- Computer Graphics and Visualization Techniques
- Language, Metaphor, and Cognition
- Remote Sensing and Land Use
- Educational Reforms and Innovations
- Advanced Vision and Imaging
- Image Enhancement Techniques
Beijing Academy of Artificial Intelligence
2023
University of Chinese Academy of Sciences
2023
Shandong Institute of Automation
2023
Chinese Academy of Sciences
2023
Télécom Paris
2021-2022
Anhui University of Technology
2022
Laboratoire Traitement et Communication de l’Information
2021-2022
InterDigital (United States)
2021
Technicolor (France)
2020
High quality facial image editing is a challenging problem in the movie post-production industry, requiring high degree of control and identity preservation. Previous works that attempt to tackle this may suffer from entanglement attributes loss person's identity. Furthermore, many algorithms are limited certain task. To these limitations, we propose edit via latent space StyleGAN generator, by training dedicated transformation network incorporating explicit disentanglement preservation...
Face age editing has become a crucial task in film post-production, and is also becoming popular for general purpose photography. Recently, adversarial training produced some of the most visually impressive results image manipulation, including face aging/de-aging task. In spite considerable progress, current methods often present visual artifacts can only deal with low-resolution images. order to achieve high quality robustness necessary wider use, these problems need be addressed. This...
Complex Query Answering (CQA) is a challenge task of Knowledge Graph (KG). Due to the incompleteness KGs, query embedding (QE) methods have been proposed encode queries and entities into same space, treat logical operators as neural set obtain answers. However, these train KG embeddings concurrently on both simple (one-hop) complex (multi-hop logical) queries, which causes performance degradation low training efficiency. In this paper, we propose Triple (Q2T), novel approach that decouples...
Generative adversarial networks (GANs) have proven to be surprisingly efficient for image editing by inverting and manipulating the latent code corresponding an input real image. This property emerges from disentangled nature of space. In this paper, we identify that facial attribute disentanglement is not optimal, thus relying on linear separation flawed. We propose improve semantic with supervision. Our method consists in learning a proxy representation using normalizing flows, show leads...
Recent work has demonstrated the great potential of image editing in latent space powerful deep generative models such as StyleGAN. However, success methods relies on assumption that a linear hyperplane may separate into two subspaces for binary attribute. In this work, we show hypothesis is significant limitation and propose to learn non-linear, regularized identity-preserving transformation leads more accurate disentangled manipulations facial attributes.
Complex Query Answering (CQA) is a challenge task of Knowledge Graph (KG). Due to the incompleteness KGs, query embedding (QE) methods have been proposed encode queries and entities into same space, treat logical operators as neural set obtain answers. However, these train KG embeddings concurrently on both simple (one-hop) complex (multi-hop logical) queries, which causes performance degradation low training efficiency. In this paper, we propose Triple (Q2T), novel approach that decouples...