Samples of the ALEGORIA-DB dataset
The ALEGORIA-DB benchmark is an image dataset involving heterogeneous geographical heritage images of various objects of interest in French urban and natural scenes, through a time period ranging from the 1920s to nowadays. The content is highly characterized by multi-date, multi-source and multi-view images. It is designed mainly for CBIR, but can also be used in related tasks:
- Cross-view image matching
- Invariant representation learning
- Few-shot landmark recognition
- Multi-temporal image matching
- Image-based geolocalization
ALEGORIA-DB consists of a total of 13175 images of high resolution (800px*variable), decomposed as follows:
- 1859 query images dispatched in 58 classes, where each class is defined around an object or a location in urban or natural scenery
- 11316 relevant image distractors
- Full statistics on the dataset can be found here: Description in English / Description en Français
Annotated variations: one originality of the dataset is that each query image was manually annotated with several quantized attributes associated with image variations, making it relevant for the evaluation of approaches facing these variations. These variations are:
Download and use rights
These data can be downloaded through a zip archive:
Contact us (here) to get the password to be able to read the archive. Please include the following informations so we can treat requests efficiently:
- Subject: "ALEGORIA-DB benchmark download"
- Message: State your name and institution
These data are made available for research purposes only. For more details on them and their conditions of use, please read the following document carefully: ALEGORIA_README.md
D. Gominski, V. Gouet-Brunet, and L. Chen, “Connecting Images through Sources: Exploring Low-Data, Heterogeneous Instance Retrieval,” Remote Sensing, vol. 13, no. 16, Art. no. 16, Jan. 2021, doi: 10.3390/rs13163080.
Date of publication: 28 December 2021.