Dr. Jian Zhang is an Assistant Professor (Ph.D. Supervisor) at the School of Electronic and Computer Engineering (SECE), Shenzhen Graduate School, Peking University, Shenzhen, China. He is now leading the Visual-Information Intelligent Learning LAB (VILLA).

He received the B.S. degree from the Department of Mathematics, Harbin Institute of Technology (HIT), Harbin, China, in 2007, and received his M.Eng. and Ph.D. degrees (under the supervision of Prof. Debin Zhao) from the School of Computer Science and Technology, HIT, in 2009 and 2014, respectively. From 2014 to 2018, he worked as a postdoctoral researcher at Peking University (PKU) (cooperated with Prof. Wen Gao) and King Abdullah University of Science and Technology (KAUST) (cooperated with Prof. Bernard Ghanem).

His research interest focuses on intelligent multimedia processing, including low-level vision, AI-generated content (AIGC) and security. He has published over 100 technical articles in refereed international journals and proceedings, with more than 7060 citations [Google Scholar]. In teaching, he won the first prize in the 23rd Young Teachers’ Teaching Basic Skills Competition at Peking University (as well as the best teaching demonstration prize and the most popular prize among students), the honorary title of “Excellent Class Teacher” at Peking University. In research, he won the 2011 Best Paper Award of IEEE International Conference on Visual Communication and Image Processing (VCIP), the 2015 Best Student Paper Award of the International Conference, the 2018 Best Paper Award of IEEE MultiMedia, and the ChinaMM 2021 Best Paper Award, Shenzhen Science and Technology Association, won the 2021 Excellent Natural Science Academic Paper Award (16 in total, the only one in the information field) and the 2023 China Artificial Intelligence Society-Huawei MindSpore Academic Award Fund Project Excellence Award. With ByteDance, he won the “NTIRE 2023 Global Challenge Panoramic Image Super-resolution Track” championship and was selected as the “Top 2% Scientists in the World” list by Stanford University for four consecutive years. In service, he is the director of the Youth Working Committee of Shenzhen Artificial Intelligence Society, the director of Guangdong Image and Graphics Society, the member of the Youth Working Committee of China Image and Graphics Society, and the executive member of the Seminar for Young Scholars in Vision and Learning (VALSE). At the same time, he serves on the editorial board of international journals such as Journal of Visual Communication and Image Representation, Signal, Image and Video Processing, CAAI Transactions on Intelligence Technology, etc.

The representative works mainly include: 1) GSR puts forward the sparse representation model and theory of image structure groups, which broke through the limitation that the previous block sparse representation model was only partially dependent, and was selected as a highly cited paper of ESI, with more than 740 citations, and was rated as one of the representative works based on sparse representation in the field of image reconstruction by many experts and scholars; 2) ISTA-Net puts forward the theory of using structured deep neural network to solve the optimization reconstruction problem, and designs a deep unfolding network inspired by iterative soft threshold algorithm, which is both interpretable and structural, greatly improving the efficiency and performance of image reconstruction, and has become the most representative work in this field, with more than 1020 citations; 3) The representative work T2I-Adapter won more than 2500 stars on GitHub less than half a year after its launch and has been combined by Stability AI, a unicorn company in the AIGC field, with its flagship model StableDiffusionXL, and used in the graffiti generation product Stable Doodle. In cooperation with Hugging Face, T2I-Adapter-SDXL was jointly developed and launched in the community.

张健,北京大学深圳研究生院信息工程学院助理教授/研究员、博士生导师,视觉信息智能学习实验室(VILLA)负责人,深圳市海外高层次人才。分别于2007年、2009年、2014年获得哈尔滨工业大学(HIT)数学与应用数学理学学士、计算机科学与技术工学硕士及计算机应用工学博士(导师:赵德斌教授)。2014-2018年期间先后在北京大学(PKU)(合作导师:高文院士)和沙特国王科技大学(KAUST)(合作导师:Bernard Ghanem教授,AI中心主任)做博士后访问研究。

主要从事基于AI的图像重建与生成,包括底层视觉与计算成像、AI内容生成(AIGC)与安全等,共计发表包括TPAMI、IJCV、TIP、CVPR、ECCV、ICCV、NeurIPS、ICLR等高水平权威国际期刊/会议论文100余篇,其中以第一/通讯作者在ACM/IEEE汇刊和CCF-A类期刊/会议上发表论文60余篇,个人Google Scholar引用7065次(其中单篇一作会议和期刊论文最高引用分别为1020次和749次),h-index值为41。入职北京大学后,于2019年创立了视觉信息智能学习实验室(VILLA)。近两年,带领VILLA以第一或通讯作者在CCF-A类会议/期刊上发表论文20余篇。相关研究成果申请/授权中国专利10余项。主持国家自然科学基金青年科学基金项目、国家自然科学基金面上项目、国家自然科学基金重点项目子课题、深圳市科技攻关项目子课题以及与字节/华为/OPPO/腾讯/创维/兔展等知名企业学术合作项目共10余项。

教学方面,获得北京大学第二十三届青年教师教学基本功比赛一等奖(以及最佳教学演示奖和最受学生欢迎奖)、北京大学“优秀班主任”、北京大学“优秀共产党员”荣誉称号等;科研方面,获得IEEE视觉通讯与图像处理(VCIP)国际会议2011年度最佳论文奖以及该国际会议2015年度最佳学生论文奖、IEEE多媒体IEEE MultiMedia国际期刊2018年度最佳论文奖、中国多媒体大会ChinaMM 2021年度最佳论文奖、深圳市科学技术协会2021年优秀自然科学学术论文奖(共16篇,信息领域唯一1篇)、2023中国人工智能学会—华为MindSpore学术奖励基金项目优秀奖,携手字节获“NTIRE 2023全球挑战赛全景图像超分辨率赛道”总冠军、连续四年入选斯坦福大学评选“全球前2%顶尖科学家”榜单等。 服务方面,担任深圳市人工智能学会青年工作委员会主任、广东省图象图形学会理事、中国图象图形学学会青年工作委员会委员、视觉与学习青年学者研讨会(VALSE)执行委员等;同时担任Journal of Visual Communication and Image Representation、Signal, Image and Video Processing、CAAI Transactions on Intelligence Technology等国际期刊编委。

代表工作主要有:1)GSR提出了图像结构组稀疏表示模型与理论,突破了前期块稀疏表示模型仅局部依赖的限制,入选ESI高被引论文,单篇引用超过740次,被多位专家学者评为图像重建领域基于稀疏表示的代表性工作之一;2)ISTA-Net提出利用结构化深度神经网络求解优化重建问题理论,设计迭代软阈值算法启发的深度展开网络,兼具可解释性和结构性,大幅提升图像重建效率和性能,已成为该领域最具代表性工作,单篇引用超过1020次;3)T2I-Adapter推出不到半年在GitHub获超过2500 stars,已被AIGC领域独角兽公司Stability AI与其旗舰模型StableDiffusionXL结合,用于涂鸦生成产品 Stable Doodle 中,服务全球用户;团队与Hugging Face公司合作开发推出了 T2I-Adapter-SDXL 在社区上线。

欢迎优秀的本科生和硕士生保送和报考北京大学信息工程学院的硕士和博士研究生,课题组博士后职位也已开放,更多最新信息请查看个人学术主页: https://jianzhang.tech/ 或者信息工程学院教师 [中文主页]以及实验室主页: https://villa.jianzhang.tech/

Research Areas

  • AI-based low-level vision and computational imaging
  • AI-generated content (AIGC) and security

Awards

Publication