Publications

More Publications

. Unsupervised Discovery of Object Radiance Fields. Preprint, 2021. PDF Website Video Code Supp
. Neural Radiance Flow for 4D View Synthesis and Video Processing. Preprint, 2021. PDF Website
. OpenRooms: An End-to-End Open Framework for Photorealistic Indoor Scene Datasets. CVPR (oral), 2021. PDF Website Short video Long video
. Letter-level Online Writer Identification. IJCV (to appear), 2020.
. Weakly supervised discriminative feature learning with state information for person identification. CVPR, 2020. PDF Code
. Unsupervised Person Re-identification by Soft Multilabel Learning. CVPR (oral), 2019. PDF Code Supp
. Patch Based Discriminative Feature Learning for Unsupervised Person Re-identification. CVPR, 2019. PDF Code
. Unsupervised Person Re-identification by Deep Asymmetric Metric Embedding. TPAMI, 2018. PDF Code
. MIXGAN: Learning Concepts from Different Domains for Mixture Generation. IJCAI, 2018. PDF
. Adversarial Attribute-Image Person Re-identification. IJCAI, 2018. PDF

Selected (Master-phase) Publications

Weakly supervised discriminative feature learning with state information for person identification

We propose a simple pseudo label model which uses state information (e.g. camera view labels in re-id and pose labels in face recognition) to refine pseudo labels. Despite simplicity, it achieves comparable performances to standard supervised fine-tuning on several datasets.
CVPR, 2020.
PDF Code

Unsupervised Person Re-identification by Soft Multilabel Learning

Lacking pairwise labels is the fundamental challenge in unsupervised person re-identification. We propose a novel paradigm to learn soft labels that are interpretable and lead to superior performances. Tackling this fundamental issue allows future research to ultimately approach supervised performances.
CVPR (oral), 2019.
PDF Code Supp

Cross-view Asymmetric Metric Learning for Unsupervised Person Re-identification

We demonstrate the potential of unsupervised person re-identification by introducing a key principle: camera view-aware modeling. This principle is now widely adopted in unsupervised/domain adaptive RE-ID works.
ICCV, 2017.
PDF Code

Selected Awards

  • Qualcomm Innovation Fellowship, by Qualcomm, 2021-2022
  • School of Engineering Fellowship, by Stanford University, 2020-2021
  • Outstanding Graduate of Guangdong Province (1⁄391), by Department of Education of Guangdong Province, 2020
  • Chinese National Scholarship (1⁄248), by Minister of Education of China, 2019
  • Chinese National Scholarship (1⁄248), by Minister of Education of China, 2018
  • Chinese National Scholarship (2⁄105), by Minister of Education of China, 2016

Service

  • Conference reviewer for CVPR18, CVPR20, NeurIPS20, WACV21, CVPR21, ICML21, ICCV21, NeurIPS21, ICLR21
  • Journal reviewer for TPAMI, Neurocomputing
  • Aiming for better reproducibility, I maintain open-source codes for all my first-authored work. Please check my github page at https://github.com/KovenYu