Weihao Xia

I am currently a PhD (MPhil) student at the the Department of Statistical Science, University College London (UCL), UK. My research topic is computer vision, specifically in controllable, interpretable, and generalizable visual contents creation. I received my Master's degree from Tsinghua University.

Email  /  Google Scholar  /  Github

Research interests

Visual Content Creation: [CVPR'21], [Neurocomputing'21], [MM'20], [NN'20]

Human Perception and Understanding: [CVPRW'21], [TCSVT'21]

There are also some paper collections based on my personal interests: image translationGitHub stars, gan inversionGitHub stars, neural renderingGitHub stars, clothed people digitalizationGitHub stars.


[04/2021] One paper is accepted to CVPR NTIRE Workshop 2021.

[03/2021] One paper is accepted to CVPR 2021.

[02/2021] One paper is accepted to Neurocomputing.

[01/2021] Our survey on GAN Inversion is available.

---- show more ----
PontTuset GAN Inversion: A Survey.
Weihao, Yulun Zhang, Yujiu Yang, Jing-Hao Xue, Bolei Zhou, Ming-Hsuan Yang.
arXiv.2101.05278 preprint

Paper / Project GitHub stars

a comprehensive overview of GAN inversion methods with an emphasis on algorithms and applications.

PontTuset Towards Open-World Text-Guided Face Image Generation and Manipulation.
Weihao Xia, Yujiu Yang, Jing-Hao Xue, Baoyuan Wu.
arXiv.2104.08910 preprint

Paper / Code GitHub stars

we propose a unified framework for both face image generation and manipulation that produces diverse and high-quality images with an unprecedented resolution at 1024 from multimodal inputs. Our method supports open-world scenarios, including both image and text.

PontTuset Region-Adaptive Deformable Network for Image Quality Assessment.
Shuwei Shi*, Qingyan Bai*, Mingdeng Cao, Weihao Xia, Jiahao Wang, Yifan Chen, Yujiu Yang.
CVPR NTIRE Workshop, 2021

Paper / Code GitHub stars

we propose the reference-oriented deformable convolution by adaptively considering the low tolerance of GAN-based distortion to spatial misalignment.

PontTuset TediGAN: Text-Guided Diverse Face Image Generation and Manipulation.
Weihao, Yujiu Yang, Jing-Hao Xue, Baoyuan Wu.
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021.

Paper / Video / Project / Code / Data GitHub stars

a novel method that unifies two different tasks (text-guided image generation and manipulation) into the same framework and achieves high accessibility, diversity, controllability, and accurateness for facial image generation and manipulation.

PontTuset Cali-Sketch: Stroke Calibration and Completion for High-Quality Face Image Generation from Human-Like Sketches.
Weihao Xia, Yujiu Yang, Jing-Hao Xue.
Elsevier Neurocomputing (JCR Q1, IF = 5.719), 2021.

Paper / Project

we propose a two-stage generative adversarial network to realize face photo synthesis from human-like sketches. It explicitly models stroke calibration and image generation using two constituent GANs. (This work is done in Apr. 2019.)

PontTuset Unsupervised Multi-Domain Multimodal Image-to-Image Translation with Explicit Domain-Constrained Disentanglement.
Weihao Xia, Yujiu Yang, Jing-Hao Xue.
Elsevier Neural Networks (NN, JCR Q1, IF= 8.05), 2020.

Paper / Project / Code

explicit disentanglement learning constraints with domain supervision to learn explicit disentangled representations and avoid the confusion of content and style.


Domain Fingerprints for No-reference Image Quality Assessment
Weihao Xia, Yujiu Yang, Jing-Hao Xue, Jing Xiao.
IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT, JCR Q1, IF = 4.685), 2020.

Paper / Project

we introduce the concept of domain fingerprint to the NR-IQA field, which is learned from image collections of different degradations and then used as the unique characteristics to identify the degradation sources and assess the quality of the image.

PontTuset InterpGaze: Controllable Continuous Gaze Redirection.
Weihao Xia, Yujiu Yang, Jing-Hao Xue, Wensen Feng.
ACM Multimedia (ACM-MM, acceptance rate 27.8%), 2020.

Paper / Project / Code and Data Data GitHub stars

we present a novel method that works on both precise redirection and continuous interpolation. With the well-disentangled and hierarchically-organized latent space, we can adjust the order and strength of each attribute by altering the additional control vector.

Updated August 2021.

Special thanks to Jon Barron for website template.