Presentation
UNO is a dataset allowing deep-learning networks to localize non-natural objects within underwater images.
This dataset is composed of five balanced training and validation sets, updated from
TrashCan, for nested cross-validation.
A test set, taken from
AquaLoc and presenting a significant domain shift, is also available for covariate shift testing.
The technical details are presented in the following paper, published in 2022 in the ICPR workshop CVAUI:
From TrashCan to UNO: Deriving an Underwater Image Dataset To Get a More Consistent and Balanced Version
Cyril Barrelet, Marc Chaumont, Gérard Subsol, Vincent Creuze, Marc Gouttefarde
5th Workshop on Computer Vision for Analysis of Underwater Imagery (CVAUI), International Conference on Pattern Recognition (ICPR), Montréal, Canada, August 2022.
Dataset details
Size |
Train/Validation |
Test |
835 Mo
|
Videos |
279 |
3 |
Frames |
5930 |
154 |
Labels |
10809 |
451 |
Train/validation examples
Test examples
Contact
If you have any question, feel free to contact us at:
cyril.barrelet@lirmm.fr