UCO-LAEO: an annotated database for training and evaluating LAEO models

Description

We use four popular TV shows: ‘Game of thrones’, ‘Mr Robot’, ‘Silicon Valley’ and ‘The walking dead’. From these shows, we collect 129 (3-12 seconds long) shots and first annotate all the heads in each frame with bounding boxes, and then annotate each head pair as LAEO or not-LAEO.

If you use this dataset in your work, please, cite [1][2].

Download

The extracted frames and annotations are available at the following URL.
To download the videos, use the following URL.

Alternatively, you can download and extract the videos, annotations and frames using:

curl http://rabinf24.uco.es/ucolaeo/ucolaeodb_v1.1.tgz | tar xz
curl http://rabinf24.uco.es/ucolaeo/ucolaeodb_videos_v1.1.tgz | tar xz

Annotations

The dataset contains 3 types of annotations:

Frame-level LAEO: is there any pair of people LAEO?
Head bounding-boxes: for each frame and each visible head, bouding boxes are provided.
Pair-level LAEO: for each frame, those pairs of heads that are LAEO are indicated.

WARNING: all the annotations fit the provided frames. Therefore, there might exist some misalignment with respect to the video files.

We consider the following set of LAEO annotations:

0: no LAEO.
1 (or 2): positive LAEO, where label 2 means that there are two people LAEO but the face of, at least, one of them is not clearly visible (e.g. occlusion).
9: ignore (due to uncertainty).

Code

After downloading the package containing the videos and annotations, within the code directory, run ucolaeo_demo.py for a quick example.

Evaluation protocol

The file code/uco_dbconfig.py contains information about the videos used for validation and testing at [1][2], the remaining ones were used for training.

References

[1]

@inproceedings{marin21pami,
  author    = {Mar\'in-Jim\'enez, Manuel J. and Kalogeiton, Vicky and Medina-Su\'arez, Pablo and and Zisserman, Andrew},
  title     = {{LAEO-Net++}: revisiting people {Looking At Each Other} in videos},
  booktitle = TPAMI,
  year      = {2021}
}

[2]

@inproceedings{marin19cvpr,
  author    = {Mar\'in-Jim\'enez, Manuel J. and Kalogeiton, Vicky and Medina-Su\'arez, Pablo and and Zisserman, Andrew},
  title     = {{LAEO-Net}: revisiting people {Looking At Each Other} in videos},
  booktitle = CVPR,
  year      = {2019}
}

Acknowledgments

The initial version of this dataset was compiled by Rafael Fernandez during the development of his final project (IT degree) at the University of Cordoba.
Thanks to RSKothari for detecting the frames' mismatch and to Isabel Jimenez for fixing it.

AVAuco/ucolaeodb