Egocentric Dataset of the University of Barcelona – Objects (EDUB-Obj) is a dataset for object localization or segmentation composed of a total of 4912 images acquired by the wearable camera Narrative, which captures images in a passive way every 30-60 seconds.
The dataset is divided in 8 different days which capture daily life activities like shopping, eating, riding a bike, working, attending meetings, commuting to work, etc. It has been acquired by 4 different subjects (from Subject1 to Subject4), and each of them having captured 2 different days (_1 and _2).
The dataset includes both the .jpg images (in the JPEGImages folder for each day), and the ground truth object segmentations (in the Annotations folder for each day), all of them sharing the same file name in JPEGImages and Annotations.
It includes a total of 11281 different object instances and the following classes (21 different classes in total) with the corresponding number of samples (ordered from more to less instances appearing in total):
- ‘lamp’ (2299)
- ‘tvmonitor’ (1274)
- ‘hand’ (1232)
- ‘person’ (1175)
- ‘glass’ (831)
- ‘building’ (732)
- ‘face’ (565)
- ‘aircon’ (530)
- ‘sign’ (506)
- ‘cupboard’ (392)
- ‘paper’ (377)
- ‘car’ (315)
- ‘bottle’ (260)
- ‘door’ (199)
- ‘chair’ (179)
- ‘mobilephone’ (145)
- ‘window’ (138)
- ‘dish’ (65)
- ‘motorbike’ (53)
- ‘bicycle’ (10)
- ‘train’ (4)
If you use this dataset for any purpose, please, do not forget to cite the following paper:
Marc Bolaños, and Petia Radeva. «Ego-object discovery.» arXiv preprint arXiv:1504.01639 (2015).
If you have any doubt or proposal, please, do not hesitate to contact the first author:
Marc Bolaños
marc.bolanos@ub.edu