Biography:
Houwen is a researcher working on computer vision and deep learning at Microsoft Research as of 2018. Before that he was a senior engineer at Qualcomm AI Research. He received Ph.D. from NLPR, Instituation of Automation, Chinese Academy of Sciences in 2016. From 2015 to 2016, he worked as a visiting research scholar at Temple University. His research interest includes neural architecture search and design, video object tracking, segmentation and detection, video moment localization, saliency detection, etc.
-
11/2020 - Our Cream NAS work was accepted by NeurIPS2020. Code is available.
-
07/2020 - Ocean is accepted by ECCV2020. Code is available.
-
06/2020 - We released the cyclic differentiable neural architecture search algorithm, arXiv and Code. The code supports distributed training, being fast, robust, and easy reproducible.
CDARTS has integrated into Microsoft NNI AutoML platform.
-
06/2020 - We released the object-aware anchor-free tracking algorithm (Ocean), arXiv and Code. The TracKit repo provides a toolkit for visual object tracking and segmentation, achieving competitive performance on VOT-2020 benchmark.
-
04/2020 - A simple transductive video object segmentation approach, dubbed TVOS, was accepted by CVPR2020. Code was released!
-
10/2019 - A coauthor paper on video moment localization, dubbed 2D-TAN, was accepted by AAAI2020. Code was merged into Microsoft Git repo. For more details, pls refer to this article.
-
09/2019 - Our team achieved Rank #1 in HACS Temporal Action Localization Challenge at ICCV2019 workshop. Congratulations to Songyang and the team! Here is the technical report.
-
07/2019 - Our team achieved one Winner and two Runner-ups in VOT2019 challenges. Code was released.
-
03/2019 - Our first tracking paper, dubbed SiamDW, was accepted by CVPR2019 as an Oral. Code is available.
