1. Filename: Unique identifier of the advertisement this face appears on
2. Bounding Box relative X1: Left-top coordinate of a rectangle identifying the face on the page, relative to the pixel coordinates of the image from column 2 (“URLs …”) of the Master Dataset (which is related to this dataset by the unique identifier in column 1). Multiply this value by the width of the image to get the absolute x coordinate. If the ad is a multi page ad, the images from column 2 have to be horizontally concatenated first.
3. Bounding Box relative Y1: Left-top coordinate
4. Bounding Box relative X2: Right-bottom coordinate
5. Bounding Box relative Y2: Right-bottom coordinate
6. Segmentation confidence score: Confidence of the neural network algorithm that these bounding boxes represent a face.
7. Size relative: 1 = Object covers all of the ad; 0.5 = Object covers half the ad.
8. Detection class id (from Open Image V4): Class index for the detected object within the list of 600 detectable objects of the Open Image V4 dataset.
9. Detection class name (from Freebase): Freebase MID of the detected object class.
10. Detection entity name (from Freebase): Freebase MID of the detected object class – in human readable format.

Authors

Ammann, N., Knäble, M, Nadj, M., Maedche, A., Kluge, S., Gehrmann, L., Stahl, F.

Funding / Grants

DFG

License

Creative Commons Attribution 4.0 International

DOI

https://doi.org/10.7801/386