UXORD-10K: Unexploded Ordnance Recognition Dataset
Overview
Unexploded Ordnance (UXO) classification is a challenging task which is currently tackled using electromagnetic induction devices that are expensive and may require physical presence in potentially hazardous environments. Our work aims to support research for image-based UXO classification with the curation of UXORD-10K, a dataset of over 10.000 annotated images shared across 8 major UXO categories.
Description
UXORD-10K was compiled by downloading UXO imagery from online resources and in particular the Collective Awareness to Unexploded Ordnance (CAT-UXO) community [1] which encourages trained professionals and individuals worldwide to contribute their data and expertise into its UXO knowledge hub. The ordnance categories of UXORD-10K are: Landmines, Fuzes, Grenades, Aircraft Bombs, Projectiles, Submunitions, Rockets and Mortars (Figure 1).
The dataset presents several challenges mainly due to the high degree of visual similarity between categories. Subtle differences in the visual appearance of different devices place the UXO recognition task in the specialized domain of fine-grained object recognition. The dataset is split into 3 sets: training, validation (for hyperparameter tuning) and test (for evaluation) and can be used as a benchmark. UXORD-10K contains 11.807 images in total (distribution shown in Table 1). The classes are unequally represented because some devices are more commonly discovered, which reflects a real-world aspect of the problem of UXO recognition.
Additional details regarding the dataset curation process as well as our experimental results and conclusions about the challenges of image-based UXO recognition be found in our research paper “Automatic Visual Recognition of Unexploded Ordnances Using Supervised Deep Learning”, which was published in the Proceedings of ICMR 2022.
[1] https://www.cat-uxo.comCitation
If you find our work useful in your research, please cite our paper as:
ACM Ref
Georgios Begkas, Panagiotis Giannakeris, Konstantinos Ioannidis, Georgios Kalpakis, Theodora Tsikrika, Stefanos Vrochidis, and Ioannis Kompatsiaris. 2022. Automatic Visual Recognition of Unexploded Ordnances Using Supervised Deep Learning. In Proceedings of the 2022 International Conference on Multimedia Retrieval (ICMR ’22). Association for Computing Machinery, New York, NY, USA, 286–294. https://doi.org/10.1145/3512527.3531383
BibTeX
| @inproceedings{10.1145/3512527.3531383, author = {Begkas, Georgios and Giannakeris, Panagiotis and Ioannidis, Konstantinos and Kalpakis, Georgios and Tsikrika, Theodora and Vrochidis, Stefanos and Kompatsiaris, Ioannis}, title = {Automatic Visual Recognition of Unexploded Ordnances Using Supervised Deep Learning}, year = {2022}, isbn = {9781450392389}, publisher = {Association for Computing Machinery}, address = {New York, NY, USA}, url = {https://doi.org/10.1145/3512527.3531383}, doi = {10.1145/3512527.3531383}, pages = {286–294}, numpages = {9}, keywords = {convolutional neural networks, object recognition, unexploded ordnance, UXORD-10K dataset, vision transformers}, location = {Newark, NJ, USA}, series = {ICMR ’22} } |
License
Attribution-NonCommercial-NoDerivatives 4.0 International CC BY-NC-ND 4.0
Contact
Panagiotis Giannakeris: giannakeris@iti.gr
Konstantinos Ioannidis: kioannid@iti.gr