Liang Zheng

Also known as 郑良

Liang Zheng is a computer vision researcher at the Australian National University whose datasets and baselines for person re-identification are widely used, and whose group now works on generative AI. In 2015, he completed a PhD at Tsinghua University and introduced the Market-1501 dataset, and in 2017 he coauthored the Random Erasing data augmentation method. He joined ANU as a Lecturer in 2018, was named one of IEEE Intelligent Systems' AI's 10 to Watch in 2022, and was named an ARC Future Fellow in 2024. The Australian Academy of Science awarded him the Brian Anderson Medal in 2026. He is a Professor in the School of Computing at ANU and works part-time as a Staff Researcher at Canva.

Now at: Australian National University and Canva
Badges: Bridge

Timeline

2010–2015

Tsinghua PhD and Market-1501

He earns a Bachelor of Science in life science at Tsinghua University in 2010 and a PhD in electronic engineering there in 2015, supervised by Shengjin Wang. In 2015, he is first author of the ICCV paper "Scalable Person Re-Identification: A Benchmark", which introduces the Market-1501 dataset with over 32,000 annotated bounding boxes and a distractor set of over 500,000 images.
2017–2019

Random Erasing, PCB and tracking

In 2017, he coauthors Random Erasing, a data augmentation method that erases a random rectangle region of a training image, and the Part-based Convolutional Baseline (PCB) for person retrieval. In 2019, he coauthors "Towards Real-Time Multi-Object Tracking", which proposes a system that learns target detection and appearance embedding in one shared model.
2018–2021

Lecturer at ANU

After postdoctoral posts at the University of Technology Sydney with Yi Yang and at the University of Texas at San Antonio with Qi Tian, he joins ANU as a Lecturer in 2018. He begins an ARC DECRA Fellowship in 2020 and is promoted to Senior Lecturer in 2021.
2022

IEEE AI's 10 to Watch

IEEE Intelligent Systems names him one of its 2022 AI's 10 to Watch. His data-centric computer vision work improves the quality of training and validation data and predicts test data difficulty without labels, and he has introduced widely used benchmarks and baselines for object re-identification.
2024

ARC Future Fellowship

In 2024, he is named an Australian Research Council Future Fellow, with over $900,000 in funding for research on deploying computer vision models reliably in the real world, including driverless cars in dense bushfire smoke or on partially flooded roads. He is also promoted to Associate Professor at ANU in 2024.
2025

REPA-E for latent diffusion

In 2025, he coauthors REPA-E with Xingjian Leng, Jaskirat Singh and Saining Xie, among others. The training recipe tunes a VAE tokenizer and a latent diffusion transformer together through a representation-alignment loss and speeds up diffusion model training by over 17x compared with REPA.
2026

Brian Anderson Medal

In 2026, the Australian Academy of Science awards him the Brian Anderson Medal, an early-career award, for contributions to representation learning in AI. A training data augmentation method he designed with collaborators is widely used in areas such as crop disease identification, surgery robots and speech understanding.
As of October 2026

Professor at ANU and Canva

He is promoted to Professor at ANU in 2026 and works part-time as a Staff Researcher at Canva. His research areas are computer vision and generative AI.