Research Experience

University of California San Diego, Feb. 2020 – Present

  • Text-to-image
    • Collected physiological images of different human organs and their corresponding physiological descriptions on the official website of medical-artist as a dataset
    • In the future, we will try to automatically generate different pathological images according to the pathological description to realize the automatic diagnosis of diseases

Zhejiang University, Nov. 2019 – Present
Advisor: Prof. Kun Kuang

  • Stable Prediction across Unknown Environments
    • Identified the causality of features, so as to learn the features with strong causality
    • In the future, we will try to extend the feature causality discrimination from binarization to continuous domain

Zhejiang University, Mar. 2019 - Nov. 2019

  • Cross-Camera Pedestrian Tracking [Project]
    • National college students innovation and entrepreneurship training program (Top 4 selected teams out of 100+ candidate teams)
    • Realized cross-camera target re-identification, moving target detection, and target tracking in a unified framework without scene overlap
    • Extracted multi-level features of pedestrians to improve the robustness and accuracy of the model
    • Designed different algorithms for different submodules, such as re-identification, moving target detection, target tracking, to improve recognition efficiency
    • Rank@1 reached 93.7% and mAP 84.2% on Market1501; Rank@1 reached 86.4% and mAP 74.5% on DukeMTMC