Politics 101

Politics 101

This datasets constis of 10 classes, where each class represents a famous politician.
Each class name is reported in each of the following 10 folders.
The main purpose of this dataset, is to offer an evaluation dataset for face detection.
However it was contructed also with Facial Recognition in mind for the 10 politicians.
The images quality is relatively low and the size of each image is not greater than 500x500.
This poses an extra challenge for face detection and for facial recognition especially of the model on this dataset.
Each folder/class contains the images corresponding to the class , as well as a .csv file with the bounding box coordinates.
The .csv format is:

"class".csv

e.g "PUTIN.csv"

And each row has the format

class_label, left, top, width, height, confidence, image, original_img_width, orginal_img_height, face_confidence_score

e.g

"Putin" , 157, 24, 55, 73, "2q_(1).jpg", 299, 168, 1

To illustrate the output of the dataset more accurately, remember that the anotattion labels
reports the bounding box absolute coordinates, by taking origin from top left corner of image
As a result, the bounding box coordinates are found on the image by doing
For bounding box width [Wstart,Wend] = [left,left+width]
For bounding box height [Hstart, Hend] = [top , top+height]
Finally the confidence score for the face is denoted as 1 for all samples