🌱_Graduate student based in Beijing_
Unsupervised Domain Adaptation(UDA) (Jun. 2022 - Present)
- UDA based on adversarial Learning.
- UDA Based on Image Translation.
- Source-free UDA, based on pseudo-labeling and denoising self-supervised techniques.
Robustness of Deep Neural Networks in Semantic Segmentation of Fluorescence Microscopy Images (Mar. 2021 -- Jun. 2022.Dec)
- We have developed an assay that characterizes robustness of DNN models in semantic segmentation of FM images.
- Our study reveals important differences in robustness of DNN segmentation models on natural images versus FM images.
- Our study reveals new and fundamental robustness properties of DNN models, and reveals relations between corruption robustness and adversarial robustness.
Robust Source-Free Domain Adaptation for Fundus Image Segmentation. (WACV 2024) Paper Code
- We proposed a two-stage training strategy for robust domain adaptation for semantic segmentation of medical images without source data.
- We utilize a pseudo-boundary loss in the target adaptation stage and develop a new domain adaptation method PLPB. Our method models edge information and achieves good performance on Average Surface Distance metric and obtain precise boundary prediction. Without requiring source data, our method achieves comparable and sometimes higher performance than state-of-the-art source-dependent UDA methods and other SF-DA methods.
- We evaluate the efficacy of PLPB on two public fundus datasets and one open domain which are popular benchmarks for UDA tasks, demonstrating improvements on both clean and adversarial samples.
By utilizing adversarial samples, our method also demonstrates good generalization capability in the open domain. Our method is flexible and can be combined with other existing adaptation techniques
- Unsupervised Domain Adaptation (UDA)
- Semantic Segmentation
- Robustness
Graduation with Honors(top 10%) (Jun. 2021)
Wuhan University
Huawei Scholarship (Jan. 2020)
Wuhan University
Excellent Group Award(top 1) (Jul. 2019)
NUS Summer School.
National Scholarship (Jan. 2019)
Wuhan University
National Scholarship (Jan. 2018)
Wuhan University
Masters in Pattern Recognition and Intelligent System
Institute of Automation, Chinese Academy of Sciences (CASIA) - Beijing, China (Aug. 2021 - Present)
Bachelor in Information Security
Wuhan University - Wuhan, China (2017- 2021)
Relevant Courses: Machine Learning (91), Discrete Mathematics (98).
Programming Skills: Python, PyTorch, C, \LaTeX.
Chinese: Native
English: C1, IELTS: 7.5