xiyichen / face-recognition

Code for independent research "Evaluate Effects of Different Face Modification Techniques for Forensic Face Recognition" at UMD

face-recognition

cleanup.py

There was a bug in the script that downloaded images for the identities. The program somehow mixed up images of two identities within the same file. This program removes the wrong images for each identity to ensure we can build the gallery correctly. Then it removes low-quality face captures (usually background faces) whose FD_SCORE or UF_SCORE is below 0.1.

gallery.py

This program builds the gallery. It sorts the sum of the absolute values of YAW, ROLL, and PITCH coordinates for each identity and select the smallest two images as "perfect images" to build the gallery.

selecting.py

This program renders a GUI to locate the face of each row on its original image. It has two buttons: the delete button will delete this face in both of the deep_features and ultraface dataframes. The continue button will simply proceed to next face.

matching.py

This program finds a best match for each of the valid faces (either identity face or teammate face) in the gallery. It also records the position of the query identity for each vector (-1 if the query is not in the gallery).

cdf.py

This program plots the CDF curve of positions of correct identities.