Projects

MAI-Image-2

Microsoft AI Superintelligence Team (2026)
Core contributor to the development and release of MAI-Image-2, a text-to-image model for high-quality image generation.
[Announcement] [Media]

MAI-Image-2.5

Microsoft AI Superintelligence Team (2026)
Core contributor to MAI-Image-2.5, which adds editing capabilities and ranked #2 on Lmarena.
[Model Page]

Publications

Think in Strokes, Not Pixels: Process-driven Image Generation via Interleaved Reasoning

European Conference on Computer Vision (ECCV2026)
Lei Zhang, Junjiao Tian, Zhipeng Fan, Kunpeng Li, Jialiang Wang, Weifeng Chen, Markos Georgopoulos, Felix Juefei-Xu, Julian McAuley, Manling Li, Zecheng He.
[Paper] [Reference]

Directional Gradient Projection for Robust Fine-tuning of Foundation Models

International Conference on Learning Representations (ICLR25)
Chengyue Huang, Junjiao Tian, Brisa Maneechotesuwan, Shivang Chopra, Zsolt Kira.
[Paper] [Reference]

Rethinking Weight Decay for Robust Fine-Tuning of Foundation Models

Neural Information Processing Systems (NeurIPS24)
Junjiao Tian, Chengyue Huang, Zsolt Kira.
[Paper] [Code]

Diffuse, Attend, and Segment: Unsupervised Zero-Shot Segmentation using Stable Diffusion

Conference on Computer Vision and Pattern Recognition (CVPR24)
Junjiao Tian, Lavisha Aggarwal, Andrea Colaco, Zsolt Kira, Mar Gonzalez-Franco.
[Paper] [Project Page] [Code]

Fast Trainable Projection for Robust Fine-tuning

Neural Information Processing Systems (NeurIPS23)
Junjiao Tian, Yen-Cheng Liu, James Seale Smith, Zsolt Kira.
[Paper] [Code]

Trainable Projected Gradient Method for Robust Fine-Tuning

Conference on Computer Vision and Pattern Recognition (CVPR23)
Junjiao Tian, Xiaoliang Dai, Chih-Yao Ma, Zecheng He, Yen-Cheng Liu, Zsolt Kira.
[Paper] [Code]

Striking the Right Balance: Recall Loss for Semantic Segmentation

International Conference on Robotics and Automation (ICRA22)
Junjiao Tian, Niluthpol Mithun, Zach Seymour, Han-Pan Chiu, Zsolt Kira
[Paper] [Code]

A Geometric Perspective towards Neural Calibration via Sensitivity Decomposition

Neural Information Processing Systems (NeurIPS21) (Spotlight < 3%)
Junjiao Tian, Dylan Yung, Yen-Chang Hsu, Zsolt Kira
[Paper] [Code] [Project Page]

Exploring Covariate and Concept Shift for Out-of-Distribution Detection

NeurIPS Workshop on Distribution Shift (NeurIPS21)
Junjiao Tian, Yen-Chang Hsu, Yilin Shen, Hongxia Jin, Zsolt Kira
[Paper] [Project Page]

Posterior Re-calibration for Imbalanced Datasets

Neural Information Processing Systems (NeurIPS20)
Junjiao Tian, Yen-Cheng Liu, Nathan Glaser, Yen-Chang Hsu, Zsolt Kira
[Paper] [Code]

When2com: Multi-Agent Perception via Communication Graph Grouping

IEEE Conference on Computer Vision and Pattern Recognition (CVPR20)
Yen-Cheng Liu, Junjiao Tian*, Nathaniel Glaser*, Zsolt Kira (* indicates equal contribution)
[Paper][Project Page]

UNO: Uncertainty-aware Noisy-Or Multimodal Fusion for Unanticipated Input Degradation

International Conference on Robotics and Automation (ICRA20)
Junjiao Tian, Wesley Cheung, Nathan Glaser, Yen-Cheng Liu, Zsolt Kira
[Paper] [Code]

Who2com: Collaborative Perception via Learnable Handshake Communication

International Conference on Robotics and Automation (ICRA20)
Yen-Cheng Liu, Junjiao Tian, Chih-Yao Ma, Nathaniel Glaser, Chia-Wen Kuo, Zsolt Kira
[Paper][Project Page]

Image Captioning with Compositional Neural Module Networks

International Joint Conference on Artificial Intelligence (IJCAI '19)
Junjiao Tian, Jean Oh
[Paper]

Junjiao Tian

Hi, I am Junjiao Tian!