wikifaces 1.0.8

Description:

wikifaces 1.0.8

Wiki Faces:



Figure 1: Joko Widodo's Wikipedia page, which includes am image of his face. The cropped image on the right is download into a directory named "Joko_Widodo."

TLDR
This project downloads images from a Wiki that include human faces. Specifically, images that are associated with
certain wikipedia categories.
Installation

Pip Installation Procedure:
From PIP:
pip install wikifaces

From Repo:
git clone [email protected]:tford9/Wiki-Faces-Downloader.git
cd Wiki-Faces-Downloader
python setup.py
pip install wikifaces

Usage
Command-Line Example
python downloader -i "indonesian engineers" -o ../data/ -d

Package Example
from wikifaces.downloader import WikiFace

wikiface_obj = WikiFace()
wikiface_obj.download(categories=['facebook'], depth=2, output_location='../data/')

The following structure is output:

facebook

cached_1_people_pages_d2.pkl
cached_pages_d2.pkl
alan_rushbridger

Alan_Rusbridger_01.jpg-p0.jpg


...
mark_zuckerberg

MarkZuckerbergcrop.jpg-p1.jpg


...



The process is carried out as follows:

Given a category from a Wiki, collect n pages that contain the same category as well as at least one category
containing "people" in the title.
With those pages, crawl across their included categories and collect y pages that contain those categories as well
as at least one "people" category.
Given the collected Wiki pages, download the primary image from the page and determine if it is a human face using
light facial detection.
We capture all images from the wiki that contain the name of the page (if it's a person then the filename contains
their name),
Using the captured name and images, we create a dataset for that face.

TODOs:

Currently, a part of this process uses a recursive call structure to get all related pages; there may be a way to
linearize, or parallelize this.
Currently, we are only pulling images contain the person's name in the title and only have one visible face in the
image. All other images are not considered. A voting system should be added to get the most represented faces across
multiple images.