|
I am a second-year Master's student at the School of Computer Science, Shanghai Jiao Tong University (SJTU), supervised by Prof. Yulun Zhang and Prof. Yutong Liu. Prior to that, I received my B.Sc. and B.Mgt. degrees from Zhejiang University (ZJU).
Google Scholar
/
Github
/
DBLP
|
|
|
[2026.07] Paper HonestFace appears at ACM MM 2026. |
|
*: Equal Contribution, †: Corresponding Author |
  
|
Jingkai Wang*, Yixin Tang*, Jue Gong, Jiatong Li, Shu Li, Libo Liu, Jianliang Lan, Yutong Liu†, Yulun Zhang† European Conference on Computer Vision (ECCV), 2026 paper / project page / code Proposed StrSR, a one-step adversarial distillation framework for DiT based real-world image super-resolution. An asymmetric discriminative distillation architecture bridges the trajectory gap of existing one-step distillation methods, and a frequency distribution matching strategy suppresses the severe grid-like periodic artifacts caused by high-frequency spectral leakage, achieving SOTA performance in Real-ISR. |
|
Jingkai Wang, Wu Miao, Jue Gong, Zheng Chen, Xing Liu, Hong Gu, Yutong Liu†, Yulun Zhang† ACM International Conference on Multimedia (ACM MM), 2026 paper / project page / code Proposed HonestFace, a multi-reference one-step diffusion model for honest face restoration under severe real-world degradation. It combines an identity embedder and masked face alignment to preserve identity-critical details and natural textures. It introduces MultiRefCeleb-Test, a large-scale real-world reference-based benchmark with 400+ identities and 9,000+ images. |
|
Light Up Your Face
|
Jue Gong*, Zihan Zhou*, Jingkai Wang, Xiaohong Liu, Yulun Zhang†, Xiaokang Yang International Conference on Machine Learning (ICML), 2026 paper / project page / code Introduce LightYourFace-160K (LYF-160K), a large-scale paired dataset built with a physically consistent renderer that injects a disk-shaped area fill light controlled by six disentangled factors. Pretrain a physics-aware lighting prompt (PALP) embedding the 6D parameters into conditioning tokens, and propose FiLitDiff, an efficient one-step diffusion model conditioned on physically grounded lighting codes that brightens underexposed faces while preserving the original scene illumination and background. |
|
Jingkai Wang, Jue Gong, Zheng Chen, Kai Liu, Jiatong Li, Yulun Zhang†, Radu Timofte, et al. New Trends in Image Restoration and Enhancement Workshop and Challenges (NTIRE, CVPR Workshop), 2026 paper / project page / code The challenge aims to generate natural, identity-consistent faces from degraded inputs, with 96 registrants, 10 teams submitting valid models, and 9 teams achieving valid final scores. |
 
|
Jue Gong*, Tingyu Yang*, Jingkai Wang, Zheng Chen, Xing Liu, Hong Gu, Yulun Zhang†, Xiaokang Yang Conference on Neural Information Processing Systems (NeurIPS), 2025 paper / project page / code Proposed a triple-branch dual-prompt guidance that leverages high-quality images, residual noise, and human motion blur masks, producing a positive–negative prompt pair for classifier-free guidance in one-step diffusion. The proposed model HAODiff achieves excellent performance in human body restoration, with significant improvements on issues such as local motion blur and clothing texture distortion. |
|
Zheng Chen, Jingkai Wang, Kai Liu, Jue Gong, Lei Sun, Zongwei Wu, Radu Timofte, Yulun Zhang†, et al. New Trends in Image Restoration and Enhancement Workshop and Challenges (NTIRE, CVPR Workshop), 2025 paper / project page / code The challenge aims to generate natural, identity-consistent faces from degraded inputs, with 141 registrants, 13 teams submitting valid models, and 10 teams achieving valid final scores. |
|
|
Zheng Chen, Kai Liu, Jue Gong, Jingkai Wang, Lei Sun, Zongwei Wu, Radu Timofte, Yulun Zhang†, et al. New Trends in Image Restoration and Enhancement Workshop and Challenges (NTIRE, CVPR Workshop), 2025 paper / project page / code The challenge aims to restore high-resolution images from bicubic x4 downsampled low-resolution inputs, attracting 286 participants and resulting in 25 valid team submissions. |
|
Jue Gong*, Jingkai Wang*, Zheng Chen, Xing Liu, Hong Gu, Yulun Zhang†, and Xiaokang Yang International Conference on Machine Learning (ICML), 2025 paper / code Developed a high-quality human body restoration dataset, PERSONA, using an automated cropping and filtering pipeline HQ-ACF. Introduced OSDHuman, a one-step diffusion model for human body restoration, which significantly outperforms existing methods in both visual quality and metrics. Contributed to the advancement of image restoration by creating a dataset that exceeds other human-related datasets in both quality and content richness. |
|
Jingkai Wang*, Jue Gong*, Lin Zhang, Zheng Chen, Xing Liu, Hong Gu, Yutong Liu†, Yulun Zhang†, and Xiaokang Yang IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025 paper / project page / code We present OSDFace, a novel one-step diffusion model for face restoration. By combining a VQ-style visual embedder, GAN discriminator, and identity loss, it effectively captures prior information. OSDFace delivers SOTA results in both qualitative and quantitative evaluations, producing high-fidelity, natural face images with strong identity consistency. |
|
Haoran Sun, Jingkai Wang, Hujun Bao, Jin Huang† ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques in Asia (SIGGRAPH Asia), 2024 paper / code Proposed a differentiable winding number framework via Gaussian kernel convolution, enabling robust inside/outside queries for an even possibly intersected 2D polygon. The differentiable formulation supports applications including intersection resolution, feature-aware curve editing with in-out brushes, and topology-preserving shape offsetting. |
|
IEEE Transactions on Image Processing (TIP)
ACM International Conference on Multimedia (ACM MM), 2026 |
|
[2026.05] ICML Gold Reviewer Award |
|
[2025.09-2026.01] Introduction to Artificial Intelligence (AI1003, Undergraduate) |
|
[2026.01-2026.05] Xiaohongshu, hilab |
|
Template from JonBarron |