README.md

metadata

license: bigscience-openrail-m

Face verification

import os import cv2 from insightface.app import FaceAnalysis import torch

prompt: compare face embediggs

class FaceRec:
    def __init__(self):
        self.foldername = '/home/emmanuel/Pictures/Webcam'
        self.files = []
        self.embeds = []
        self.diff = []
        self.ground_mathches = []
        self.sampling = None
        
            

    def folder(self, attempt=True, folder='/home/emmanuel/Pictures/Webcam'):
        if attempt:
            for file in os.listdir(folder):
                self.files.append(file)

            self.image_pair = list(zip(self.files[0:len(self.files)//2], self.files[len(self.files)//2:]))
            print(self.image_pair)


        else:
            self.foldername = '/home/emmanuel/Pictures/webcam'
            self.files = []
            self.folder(attempt=True, folder=self.foldername)
            

        
    
    def embeddings(self, image):
        app = FaceAnalysis(name="buffalo_l", providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
        app.prepare(ctx_id=0, det_size=(640, 640))
        image1 = cv2.imread(image)
        faces = app.get(image1)

        faceid_embeds = torch.from_numpy(faces[0].normed_embedding).unsqueeze(0)
        return(torch.Tensor(faceid_embeds))



    def face_embed(self, face, face1):
        # Load the two images and get their face embeddings.
        face_encodings = self.embeddings(face)
        face_encodings1 = self.embeddings(face1)
        return(torch.nn.functional.cosine_similarity(face_encodings, face_encodings1))
    


    def closeness(self):
        self.embeds = []
        for faces in self.image_pair:
            self.embeds.append(self.face_embed(self.foldername+'/'+faces[0], self.foldername+'/'+faces[1]))

        return(0)
    

    def compare(self, attempt=True):
        self.diff = []
        for diffs in list(zip(self.embeds[0:len(self.embeds)//2], self.embeds[len(self.embeds)//2:])):
            self.diff.append(torch.nn.functional.pairwise_distance(diffs[0], diffs[1]))



    
    def expectation(self):
        mean, std = torch.mean(torch.Tensor(self.diff[0:])), torch.std(torch.Tensor(self.diff[0:]))
        distribute = torch.distributions.Normal(mean, std)
        self.sampling = distribute.sample(sample_shape=(10,))



    def model(self):       
        self.closeness()
        return(self.compare())



    def verify(self):
        self.folder()
        self.model()
        self.expectation()
        self.folder(attempt=False)
        self.model()

        fails = 0
        success = 0
        max_itter = 10
        while max_itter >= 0:
            for samples in self.sampling:
                if self.diff[0] <= samples:
                    success = success+1
                
                else:
                    fails = fails+1
                
            max_itter = max_itter-1


        if fails > success:
            return(False)

        else:
            return(True)


    



Recognition = FaceRec()
print(Recognition.verify())