Hi, I'm Junwen Pan.
I'm a computer vision researcher at ByteDance. My research interest includes weakly/semi-supervised image segmentation, vision-language learning, and explainable machine learning. Currently, I am working on VQA grounding.
Hybrid Supervised Segmentation
Label-efficient Hybrid-Supervised Learning for Medical Image Segmentation.
(AAAI 2022, TPAMI in progress)
Junwen Pan, Qi Bi , Yanzhan Yang, Pengfei Zhu and Cheng Bian
Weakly and Semi-Supervised Segmentation
Learning Self-Supervised Low-Rank Network for Single-Stage Weakly and Semi-Supervised Semantic Segmentation. (IJCV 2022)
Junwen Pan, Pengfei Zhu, and et. al.
Publication, arXiv, Github
ABCNet: A New Efficient 3D Dense-Structure Network for Segmentation and Analysis of Body Tissue Composition on Body-Torso-Wide CT Images. (Medical Physics 2020)
Tiange Liu, Junwen Pan, Drew A. Torigian, Pengfei Xu, Qiguang Miao, and et. al.
CVPR 2022 VizWiz Grand Challenge Workshop
Tell Me the Evidence? Dual Visual-Linguistic Interaction for Answer Grounding. (CVPR 2022 Workshop)
Our solution DaVI ranked 1st place on the answer grounding track.
Junwen Pan, Guanlin Chen, and Yi Liu.
Paper, Video, Leaderboard, Challenge Website
CVPR 2021 Learning from Limited and Imperfect Data
This challenge aims to learn pixel-level segmentation using image-level annotations.
Our SLRNet ranked 1st place on the validation set and 2nd place on the test set in the weakly supervised semantic segmentation track.
Junwen Pan, Yongjuan Ma, and Pengfei Zhu.
Poster, Slides, Challenge Website
MICCAI 2021 Gastrointestinal Image ANAlysis Challenge
This challenge involved detection, segmentation and classification in colonoscopy.
Our team ranked 1st place in the detection track and 2nd place in the segmentation track.
Junwen Pan, Shijie Liu, Xiaozhou Shi, and Hong-Yu Zhou.
Challenge Website, Result Paper
Project on Github
A simple comment system for modern blog systems, serving thousands of websites around the world. (Star 600+, Fork 400+)
I have organized two challenge workshops in conjunction with ECCV 20 and ICCV 21, which have attracted 100+ teams each year.
Tools for DL Ops
I've developed a series of scripts and tools to improve the efficiency of DL and shared them on my GitHub and blog.