📝 Publications
A full publication list is available on my google scholar page.
[Open-Source Project] FaceChain: A Playground for Human-centric Artificial Intelligence Generated Content
[ModelScope Studio]
[Alibaba Cloud API]
[HuggingFace Space]
- FaceChain is a novel framework for generating identity-preserved human portraits.
- FaceChain has both high controllability and authenticity in portrait generation, including text-to-image and inpainting based pipelines, and is seamlessly compatible with ControlNet and LoRAs.
[ICCV 2023] TransFace: Calibrating Transformer Training for Face Recognition from a Data-Centric Perspective
Jun Dan, Yang Liu, Haoyu Xie, Jiankang Deng, Haoran Xie, Xuansong Xie, Baigui Sun
- TransFace is a cutting-edge facial representation extractor in the AIGC era, designed to capture fine-grained facial features at the patch-level.
- TransFace is integrated in FaceChain as a key identity-preserved module to assist Stable Diffusion in generating human portraits with fine-grained facial details and diverse styles.
- Industry Impact: TransFace model has reached over 15K+ downloads on the ModelScope platform, and has been applied in various facial AIGC projects, such as Alibaba Tongyi Wanxiang (通义万象写真馆), FaceChain, Fliggy (飞猪数字旅拍), etc.
[NeurIPS 2024] TopoFR: A Closer Look at Topology Alignment on Face Recognition
Jun Dan, Yang Liu, Jiankang Deng, Haoyu Xie, Siyuan Li, Baigui Sun, Shan Luo
- TopoFR is the first attempt to leverage the powerful and substantial structure information hidden in large-scale face dataset to improve the generalization performance of face recognition models.
- TopoFR achieves SOTA performance on various face benchmarks.
[NeurIPS 2024] TFGDA: Exploring Topology and Feature Alignment in Semi-supervised Graph Domain Adaptation through Robust Clustering
Jun Dan, Weiming Liu, Chunfeng Xie, Hua Yu, Shunjie Dong, Yanchao Tan
- TFGDA is an advanced graph transfer learning framework that leverages the intrinsic topological structure information embedded in graphs to improve model generilization performance across domains.
- TFGDA showcases superior performance on multiple transfer learning benchmarks.
[ACM MM 2024] HOGDA: Boosting Semi-supervised Graph Domain Adaptation via High-Order Structure-Guided Adaptive Feature Alignment
Jun Dan, Weiming Liu, Mushui Liu, Chunfeng Xie, Shunjie Dong, Guofang Ma, Yanchao Tan, Jiazheng Xing
- HOGDA is a novel graph transfer learning framework that incorporates a high-order structure information mixing module, effectively capturing abundant structure information in graphs and greatly enhancing the feature extractor’s adaptability across different domains.
- HOGDA demonstrates remarkable transfer performance on various benchmarks.
[CVPR 2025] Distinguish Then Exploit: Source-free Open Set Domain Adaptation via Weight Barcode Estimation and Sparse Label Assignment
Weiming Liu*, Jun Dan*, Fan Wang, Xinting Liao, Junhao Dong, Hua Yu, Shunjie Dong, Lianyong Qi (*: equal contribution.)
- DTE is a novel framework designed for source-free open set domain adaptation problem. By integrating weight barcode estimation with sparse label assignment, DTE enables efficient and robust knowledge transfer across domains.
- DTE outperforms SOTA models on tackling the source-free open set domain adaptation problem.