DAODiS: Domain-Adapted Object Detection in Surveillance Videos

Welcome to DAODiS!!

In video surveillance applications, dealing with domain shifts and unknown classes is critical but still open issue. Recently, many object detection methods have been proposed using CNN such as YOLO, Faster-RCNN, and Retina-net . However, the performance of these methods are very sensitive against domain shifts. We aim to tackle the challenge for developing more accurate object detection with fewer annotations.


  • Domain-shift such as night scenes and thermal scenes

Important Date

This schedule is a tentative schedule. Due to COVID-19, this schedule changes.

  • Dec. 1, 2019: ad-hoc website open.
  • Feb. 5, 2020: Dataset for development available.
  • The competition at AVSS2020 has been cancelled.


  • Atsushi Shimada (Kyushu Univ.)
  • Janusz Konrad (Boston Univ.)
  • Vincent Charvillat (ENSEEIHT)
  • Tsubasa Minematsu (Kyushu Univ.)
  • Takashi Shibata (NEC Corp.)
  • Yasutomo Kawanishi (Nagoya Univ.)