gyguo/odsawl

code for "Learning Object Detectors with Semi-Annotated Weak Labels", published in TCSVT2018

Guangyu Guo, PhD (Chinese name: 郭广宇) 👋

[Google Scholar] [知乎]

About Me

• Current State: I am pursuing my Ph.D. degree at Northwestern Polytechnical University, also as a Research Intern at Medical AI Laboratory, Damo, Alibaba.
• Research Interests: Computer Vision and Artificial Intelligence related topics, especially in weakly supervised learning, knowledge distillation, medical image analysis, and vision-language model.

Education/Work Experience

• $\bullet$ 2023-Present: Research Intern at Medical AI Laboratory, Damo (达摩院) Alibaba (阿里巴巴)
• $\bullet$ 2021-2022: Inter at the third academy of China Aerospace Science and Industry Corporation (CASIC)
• $\bullet$ 2018-Present: Northwestern Polytechnical University, Ph.D, supervised by Prof. Junwei Han and Prof. Dingwen Zhang
• $\bullet$ 2016-2018: Northwestern Polytechnical University, MA, supervised by Prof. Junwei Han
• $\bullet$ 2012-2016: Nanjing University of Aeronautics and Astronautics, BA

Research Experience

Time	Domain	Papers
2021 - Present	AI for Medical	IE-Net(JBHI), PAE-ASS(Neural Netw.), RG-CD(Arxiv24) 1 TMI under review
2021 - Present	Knowledge distillation	SPU(Vis. Intell.), Pixel Distillation(TPAMI) 1 TPAMI under review
2018 - 2022	Weakly-supervised Learning	ODSAWL(TCSVT), SLT-Net(CVPR), GWSOL(TNNLS)
2016 - 2017	Flow estimation, Keypoint detection	PoseFlow(CVPR)

Project Experience

Time	Task	Projects	Source
2023 - Present	Medical Image Analysis	AI-powered multi-cancer early detection public welfare program, CT-based Survival Prediction	Medical AI Lab, Alibaba Damo
2022 - Present	Medical Image Analysis	Nuclei recognition, Lung ultrasound B-Line detection	Xijing Hospital, The Fourth Military Medical University
2021 - 2022	Object and Keypoint Detection	Infrared ship recognition under complex background	China Aerospace Science and Industry Corporation (CASIC)

News

• $\bullet$ 2024.07: One paper was accepted to TPAMI.
• $\bullet$ 2024.04: Congratulations to Kaihui Yang on obtaining the Ph.D qualification of NWPU
• $\bullet$ 2024.01: One paper was accepted to ACM TOMM. Congratulations to Kaihui Yang.
• $\bullet$ 2024.01: One paper was accepted to Neural Networks. Congratulations to Jieru Yao.

Paper List

$^{\star}$ equal contribution; 📧 corresponding author.

• $\bullet$ Guangyu Guo, Dingwen Zhang, Longfei Han, Nian Liu, Ming-Ming Cheng, Junwei Han. Pixel Distillation: Cost-flexible Distillation across Image Sizes and Heterogeneous Networks. TPAMI, 2024.
• $\bullet$ Kaihui Yang, Junwei Han, Guangyu Guo, Chaowei Fang, Yingzi Fan, Lechao Cheng, Dingwen Zhang. Progressive Adapting and Pruning: Domain-Incremental Learning for Saliency Prediction. ACM TOMM, 2024.
• $\bullet$ Jieru Yao, Longfei Han, Kaihui Yang, Guangyu Guo, Nian Liu, Xiankai Huang, Zhaohui Zheng, Dingwen Zhang, Junwei Han. Contextual Dependency Vision Transformer for spectrogram-based multivariate time series analysis. Neurocomputing, 2024.
• $\bullet$ Guangyu Guo, Jiawen Yao, Yingda Xia, Tony CW Mok, Zhilin Zheng, Junwei Han, Le Lu, Dingwen Zhang, Jian Zhou, Ling Zhang. Boosting Medical Image-based Cancer Detection via Text-guided Supervision from Reports. arXiv preprint, 2024.
• $\bullet$ Jieru Yao, Longfei Han, Guangyu Guo, Zhaohui Zheng, Runmin Cong, Xiankai Huang, Jin Ding, Kaihui Yang, Dingwen Zhang, Junwei Han. Position-based anchor optimization for point supervised dense nuclei detection. Neural Networks, 2024.
• $\bullet$ Guangyu Guo, Longfei Han, Le Wang, Dingwen Zhang, Junwei Han. Semantic-aware Knowledge Distillation with Parameter-free Feature Uniformization. Visual Intelligence, 2023.
• $\bullet$ Dingwen Zhang, Junwei Han, Guangyu Guo, Long Zhao. Generalized Weakly Supervised Object Localization. TNNLS, 2022.
• $\bullet$ Guangyu Guo, Junwei Han, Fang Wan, Dingwen Zhang. Strengthen learning tolerance for weakly supervised object localization. CVPR, 2021.
• $\bullet$ Guangyu Guo, Zhuoyan Liu, Shijie Zhao, Lei Guo, Tianming Liu. Eliminating indefiniteness of clinical spectrum for better screening COVID-19. JBHI, 2021.
• $\bullet$ Dingwen Zhang, Guangyu Guo $^{\star}$, Dong Huang, Junwei Han. Poseflow: A deep motion representation for understanding human behaviors in videos. CVPR. 2018.
• $\bullet$ Dingwen Zhang, Junwei Han, Guangyu Guo, Long Zhao. Learning object detectors with semi-annotated weak labels. TCSVT, 2018.

odsawl's Issues

ROIS offset

Why should I set the offset for rois in train.m? Can I just scale rois by 1/16 without offset?

if is_vgg16
o0 = 8.5 ;
o1 = 9.5 ;
else
o0 = 18 ;
o1 = 9.5 ;
end

rois = [ rois(1,:); ...
floor((rois(2,:) - o0 + o1) / ss(1) + 0.5) + 1;
floor((rois(3,:) - o0 + o1) / ss(2) + 0.5) + 1;
ceil((rois(4,:) - o0 - o1) / ss(1) - 0.5) + 1;
ceil((rois(5,:) - o0 - o1) / ss(2) - 0.5) + 1];