Request for Enhancement: Customizable Thresholds in Deepface.find Function

[RicaredoAndrade]

Description

I've been using Deepface for testing anonymization solutions and have found it to be a powerful tool. However, I've encountered a limitation in the Deepface.find function that I believe could significantly enhance its usability.

Issue

In the current implementation, the face recognition model thresholds are predefined and cannot be manually selected. This limitation poses challenges when computing the Cumulative Match Characteristic, particularly when comparing an image with all testing identities, as the predefined threshold restricts the computation of the rank for each identity.

Enhancement Request

I propose adding a feature to the Deepface.find function that allows users to manually define the face recognition model threshold. This enhancement would not only expedite the computation of the Cumulative Match Characteristic but also provide users with the flexibility to test different threshold values, improving overall testing capabilities.

Solution

Add a new parameter to the function

deepface/deepface/DeepFace.py

Line 185 in 88814e6

def find(

threshold = None

Add it to the return

deepface/deepface/DeepFace.py

Line 229 in 88814e6

return recognition.find(

threshold = None

Add the parameter to the function

deepface/deepface/modules/recognition.py

Line 20 in 88814e6

def find(

threshold = None

Adjust the threhsold value

deepface/deepface/modules/recognition.py

Line 254 in 88814e6

threshold = dst.findThreshold(model_name, distance_metric)

if threshold == None:
    threshold = dst.findThreshold(model_name, distance_metric)

[serengil]

Seems a fair request. Will do it in a couple of days.

[serengil]

Closed with PR - #974