Daoan Zhang

I am currently a first-year Phd student at University of Rochester, advised by Albert Arendt Hopeman Professor Jiebo Luo. I was a graduate student (thesis-based master) at Southern University of Science and Technology, advised by Prof. Jianguo Zhang. Before that, I received my bachelor's degree from East China University Of Science And Technology.

I was a research intern at Tencent AI Lab, advised by Dr. Jianhua Yao and Chenchen Qin in 2023. I was also a research intern at Ping An Technology, advised by Dr. Lingyun Huang in 2022. I remotely joined CCVL (Computational Cognition, Vision, and Learning) research group at Johns Hopkins University as an intern, advised by Prof. Alan L. Yuille.

Research

My research interests lie in the field of Large Model Training . The most of my current works focused on Computer Vision , Multi-Modality and AI for Science .

GPT-4V(ision) as A Social Media Analysis Engine
Hanjia Lyu*, Jinfa Huang*, Daoan Zhang*, Yongsheng Yu*, Xinyi Mou*, Jinsheng Pan, Zhengyuan Yang, Zhongyu Wei, Jiebo Luo
Arxiv
We explore GPT-4V(ision)'s capabilities for social multimedia analysis including sentiment analysis, hate speech detection, fake news identification, demographic inference, and political ideology detection.

DNAGPT: A Generalized Pretrained Tool for Multiple DNA Sequence Analysis Tasks
Daoan Zhang, Weitong Wang, Bing He, Jianguo Zhang, Chenchen Qin, Jianhua Yao,
Under Review by Nature Methods
We propose DNAGPT, a generalized DNA pre-training model trained on over 200 billion base pairs from all mammals. By enhancing the classic GPT model with a binary classification task (DNA sequence order), a numerical regression task (guanine-cytosine content prediction), and a comprehensive token language, DNAGPT can handle versatile DNA analy- sis tasks while processing both sequence and numerical data.

Cross Contrastive Feature Perturbation for Domain Generalization
Chenming Li, Daoan Zhang, Wenjian Huang, Jianguo Zhang,
ICCV 2023
In this paper, we propose an online one-stage Cross Contrasting Feature Perturbation (CCFP) framework to simulate domain shift by generating perturbed features in the latent space while regularizing the model prediction against domain shift.

Rethinking Alignment and Uniformity in Unsupervised Image Semantic Segmentation
Daoan Zhang, Chenming Li, Haoquan Li, Wenjian Huang, Lingyun Huang, Jianguo Zhang,
AAAI 2023 (Oral)
In this paper, we address the critical properties from the view of feature alignments and feature uniformity for UISS models. We also make a comparison between UISS and image-wise representation learning. we proposed a robust network called Semantic Attention Network(SAN).

Feature Alignment and Uniformity for Test Time Adaptation
Shuai Wang, Daoan Zhang, Zipei Yan, Jianguo Zhang, Rui Li
CVPR 2023
We first address TTA as a feature revision problem due to the domain gap between source domains and target domains. After that, we follow the two measure- ments alignment and uniformity to discuss the test time fea- ture revision. For test time feature uniformity, we propose a test time self-distillation strategy to guarantee the consis- tency of uniformity between representations of the current batch and all the previous batches. For test time feature alignment, we propose a memorized spatial local cluster- ing strategy to align the representations among the neigh- borhood samples for the upcoming batch.

Prototype Knowledge Distillation for Medical Segmentation with Missing Modality
Shuai Wang, Zipei Yan, Daoan Zhang, Haining Wei, Zhongsen Li, Rui Li
ICASSP 2023
In this paper, we propose a prototype knowledge distillation (ProtoKD) method to tackle the challenging problem, especially for the toughest scenario when only single modal data can be accessed. Specifically, our ProtoKD can not only dis- tillate the pixel-wise knowledge of multi-modality data to single-modality data but also transfer intra-class and inter-class feature variations, such that the student model could learn more robust feature representation from the teacher model and inference with only one single modal- ity data.

TransVLAD: Focusing on Locally Aggregated Descriptors for Few-Shot Learning
Haoquan Li, Laoming Zhang, Daoan Zhang, Lang Fu, Peng Yang, Jianguo Zhang,
ECCV 2022
This paper presents a transformer framework for few-shot learning, termed TransVLAD, with one focus showing the power of locally aggregated descriptors for few-shot learning.

Aggregation of Disentanglement: Reconsidering Domain Variations in Domain Generalization
Daoan Zhang, Mingkai Chen, Chenming Li, Lingyun Huang, Jianguo Zhang,
In Submission to IJCV
We proposed a new perspective to utilize class-aware domain variant features in training, and in the inference period, our model effectively maps target domains into the latent space where the known domains lie. We also designed a contrastive learning based paradigm to calculate the weights for unseen domains.

Semi-Supervised Semantic Segmentation via Boosting Uncertainty on Unlabeled Data
Daoan Zhang, Yunhao Luo, Jianguo Zhang,
Arxiv
We figure out that the distribution gap between labeled and unlabeled datasets cannot be ignored. To address this issue, we theoretically analyze and experimentally prove that appropriately boosting uncertainty on unlabeled data can help minimize the distribution gap, which benefits the generalization of the model. We propose two strategies and design an algorithm of uncertainty booster specially for semi-supervised semantic segmentation.

Services

Review Service: ICCV2023, AISTATS2024, CVPR2024, EACL2024, TCSVT, Pattern Recognition

Teaching Assistant, CS324 Deep Learning

credits