This dataset is collected by Xiaoyang Li from Zhejiang University and Kanzhi Wu from University of Technology Sydney. We would also like to thank for the generous help from Prof. Yong Liu.

RGB-D descriptor evaluation dataset

This dataset consists 4 subsets considering following variations individually:

  • Illumination: 4 levels of synthetic illumination variations and neutral-density filter for natural illumination variation
  • Rotation: 36 frames of synthetic in-plane rotation and 10 frames of 3d rotation.
  • Scale: 10 frames of scale variations by moving camera backwards every ~10 cm.
  • Viewpoint: 24 frames by moving sensor around the scene.

The dataset can be downloaded from link.

Snapshots of this dataset are below:

Illumination variation illu

Rotation variation rotation

Scale variation scale

Viewpoint variation viewpoint

RGB-D scene dataset

This dataset are collected under high illumination variations, the indoor light was turned off during the data collect.


This dataset is constructed to evaluate our new RGB-D feature, RISAS, Rotation, Illumination, Scale invariant Appearance and Shape feature. Please visit RISAS for detail information.