Dataset for HDR4CV: High dynamic range dataset with adversarial illumination for testing computer vision methods
Param Hanji, Muhammad Zeshan Alam, Nicola Giuliani, Hu Chen, Rafał Mantiuk · Apollo (University of Cambridge) · 2021
This dataset is intended to test the robustness of computer vision methods under challenging illumination conditions. The dataset includes four scenes (*Faces*, *Tunnel*, *Street-parallel*, *Street-diagonal*), which were captured multiple times, each time under a different illumination condition (**night**, **hdr**, **glare**, and **uniform**). A non-challenging **uniform** illumination condition serves as a reference and can be used to generate labels for the other three challenging illumination conditions and test a given computer vision method. This lets us avoid the tedious task of manually labeling the data. Refer to README.md for further details.