I am now working as a senior applied scientist at Amazon, Prime Video. I am broadly interested in computer vision and machine learning, with more specific focus on visual recognition and scene understanding.
Contact:
Xiang Yu (喻祥)
Email: yuxiang03@gmail.com
News:
Two ICCV'2023 papers got accepted! One is integrating the transformer architecture with Deformable Models to solve the 3D shape abstraction problem, achieving clear improvement over the SOTAs! The other is leveraging the gradient signal to noise ratio as a cue to tackle the domain generalization problem, showing consistent advantage over other stochastic approaches including DropBlock, RSC, etc.
Two CVPR'2023 papers got accepted! Congratulations to the co-authors for their excellent works!
One ICLR'2023 paper got accepted and one WACV'2023 paper got accepted! In our ICLR'23 paper, we propose a progressive mix-up algorithm to deal with the few-shot multi-source domain transfer problem, which surprisingly achieves new SOTA records than the top methods, such as MDAN, Mix-up, DAML, URT, etc. The code will be released soon!
I will be serving for the 32nd International Joint Conference on Artificial Intelligence as a Senior Program Committee member. Looking forward to interesting and high quality submissions!
I am newly joining Amazon, Prime Video as a senior applied scientist. Cherish and much appreciate the journey with NECLA previous colleagues! Now a new and exciting journey for me on the way!
Our team gets two papers accepted to the ECCV2022! One is related to better associate the language embedding to the visual embedding for Vision-Language Pre-training. The other is about privacy-aware computational photography. Congratulations to the team members!
Our team gets one Oral and two posters in the upcoming CVPR2022! Congratulations to the team members and appreciate the recognition from the conference committee!
I will be serving for Thirty-Sixth AAAI Conference on Artificial Intelligence 2022 as a SPC member. Looking forward to high quality submissions to this top-tier AI international conference!
Our video domain adaptation paper "Learning Cross-Modal Contrastive Features for Video Domain Adaptation" got accepted to ICCV2021! Congratulations to Donghyun!
Our "Cross-Domain Similarity Learning for Face Recognition in Unseen Domains" paper is accepted to CVPR 2021! A metric-level domain generalization is proposed to deal with the imbalanced training and testing data distribution problem. Congratulations to Masoud!
Our large scale unsupervised learning for deep face recognition paper got accepted to ECCV 2020! We provide a practical solution on how to leverage large scale unlabeled data to boost the recognition performance, which shows orthogonal improvement on SOTA recognition frameworks. Congratulations to Aruni!
Our "Universal Face Recognition" paper and "Private-kNN Differential Privacy" paper got accepted to CVPR 2020! Face Recognition paper strikes new state-of-the-art, congratulations to Yichun! Private-kNN paper is the first to apply differential privacy on real computer vision datasets and achieved 90% less privacy loss than state-of-the-art, congratulations to Yuqing!
Our "GLoSH: A Global-Local Spherical Harmonics for Intrinsic Image Decomposition" paper got accepted as an Oral to ICCV 2019! Congratulations to Hao!
Our "Feature Transfer Learning" paper and "Joint Pixel and Feature level Domain Adaptation" paper got accepted to CVPR 2019!
Our semi-supervised "Deep Metric Learning with Feature Transfer Learning" got accepted to ICLR 2019!
"Deep Supervision with Intermediate Concepts" article got accepted to TPAMI, 2018.