This dataset accompanies our paper titled 'Memorable Maps: A Framework for Re-defining Places in Visual Place Recognition' The PDF for this paper can be found at (https://arxiv.org/abs/1811.03529).
Unlike existing VPR datasets which provide either viewpoint variation or conditional variation, ESSEX3IN1 is the first dataset that consists of images of places that are confusing for both VPR and human-cognition. It contains confusing and challenging dynamic objects, natural scenes and low-informative frames. Most of the state-of-the-art VPR techniques struggle in catering for these confusing images as shown in our paper.
The dataset is divided into 2 folders. Frame-to-frame correspondence is applicable across query and reference frames. Images 0-132 in each combined folder are confusing while images 133-209 are good frames.
If you use this dataset in your work, please cite our paper as below.
@article{zaffar2018memorablemaps, title={Memorable Maps: A Framework for Re-defining Places in Visual Place Recognition}, author={Zaffar, Mubariz and Ehsan, Shoaib and Milford, Michael and McDonald-Maier, Klaus}, journal={https://arxiv.org/abs/1811.03529}, year={2018} }