cumulative-learning

We are releasing the ground truth annotations in MOT file format for the following datasets:

• MusicDataset: converted from the official release
• BigBangTheory: converted from the official release (original_gt files) and a version we manually annotated including all the actors not previously annotated. For each identity we included the values (#id, name, first appearance, # frames from first episode) inside the file names_to_id
• Motinas Multi Face Dataset: manually annotated Ground Truth using OpenCV CVAT tool

Citation

Pernici_2018_CVPR

@InProceedings{Pernici_2018_CVPR,
author = {Pernici, Federico and Bartoli, Federico and Bruni, Matteo and Del Bimbo, Alberto},
title = {Memory Based Online Learning of Deep Representations From Video Streams},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}
}

Video

Motinas Multi Face Dataset

ftp://spit.eecs.qmul.ac.uk/pub/multi_face

Music Dataset

https://sites.google.com/site/shunzhang876/eccv16_facetracking

BigBangTheory (ep 01-06, Season 1)

From DVD Release, see Bäuml, M., Tapaswi, M., Stiefelhagen, R., 2013. Semi-supervised Learning with Constraints for Person Identification in Multimedia Data, in: IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 3602–3609.