Day-night image matching is important for many computer vision applications ( e.g. large scale localization and place recognition). However, performance of current methods are far from satisfactory. In this project, we evaluated local features under day-night illumination changes. We found that:

  • Detectors are affected by day-night illumination changes to a large extent.
  • High repeatability of detector does not mean good matching performance.
  • There is great potential for improving both detectors and descriptors for night images.



We collect our own data set for evaluation from the AMOS data set [1]. These images are taken at different times of the day by fixed webcams. There are 17 image sequences containing 1722 images in total. We provide the time stamp for each of these images. The dataset is available for Download.


[Paper] [Bibtex] [Dataset] [Poster] [Slides]


  1. N. Jacobs, N. Roman and R. Pless. Consistent temporal variations in many outdoor scenes. In: CVPR (2007)
  2. H. Zhou, T. Sattler and D. Jacobs. Evaluating Local Features for Day-Night Matching. In: ECCV workshop (2016)