Xin Wen (温鑫)

I am a Ph.D. candidate at CVMI Lab, The University of Hong Kong, advised by Dr. Xiaojuan Qi. Prior to that, I obtained my B.Eng. degree in computer science with minor in applied mathematics at Tongji University. I also spend some time at ByteDance AI Lab, MEGVII Research, and Shanghai AI Laboratory.

My research interest mainly lies in learning robust and generalizable vision representations with minimal human intervention, and its application to recognition/perception tasks.

Please feel free to drop me an email if you are interested in what I do and seek for possible collaborations.

Email  /  Github  /  Google Scholar  /  Twitter  /  LinkedIn

We build a counterfactual visual question answering benchmark, and show that strong Vision-Language Models, even GPT-4, cannot handle them very well.

We bridge the gap between vision & language spaces by reformulating region-word alignment as co-occurring object discovery, and images mention a shared concept in their captions are grouped together.

We revisit the reason that makes previous parametric classifiers fail to recognise new classes for GCD, identify the prediction biases between and within seen and novel classes as the key issue, and propose a simple yet strong framework that addresses these limitations and achieves state-of-the-art performance in this field.

We tackle GCD without knowing the class number as a-priori, propose a semi-supervised variant of GMM with stochastic splitting and merging to dynamically determine prototypes, and leverage prototpyical contrastive learning for representation learning on partially labelled data.

We propose the Masked Scene Contrast (MSC) framework for unsupervised 3D representation learning, which efficiently generates contrastive views directly on scene-level point clouds and enables large-scale 3D pre-training across multiple datasets.

We show that object discovery can be learned jointly with the representations from scratch on real-world scene-centric data, which leads to strong transfer learning results in various downstream tasks.

We present a contrastive learning-based video representation learning framework that adopts long-range temporal information between frame-level features using self-attention.

We leverage self-supervised learning and knowledge distillation to improve the generalizability of CNN models for image classification under the data-deficient setting.

Reviewer for TPAMI, NeurIPS, ICLR, ICML, CVPR, ICCV, ECCV, WACV, CVinW, and OOD-CV.

Template gratefully stolen from here.