Face Recognition Project

Introduction

• This repo is a reimplementation of Arcface(paper), or Insightface(github)
• This repo is get inspiration from insightface, Insightface_Pytorh

SetUp

Install MTCNN FaceDetector

Follow this instruction FaceDetector Installation.

Install

python setup.py install

Basic Usage

Download the pre-trained model.

Download from pre-trained model and merge the part files

cat model_ir_se50.pth* > model_ir_se50.pth
mv model_ir_se50.pth /path/to/FaceRecognizer/output/res50

FaceVerify

Verify if they are the same person.

from insight_face import FaceSearcher

# Load pre-trained res50 model
searcher.load_state('./output/res50/model_ir_se50.pth', 50)

# Load face images
face1 = cv2.imread('./tests/assets/reba/2.jpg')
face2 = cv2.imread('./tests/assets/reba/3.jpg')

result = searcher.verify(face1, face2)

assert result is True

Search

Find who they are.

face_bank_dir = 'tests/assets/'
multi_face_img = cv2.imread("tests/assets/multi_face.jpg")

searcher.add_face_bank(face_bank_dir, force_reload=True, bank_name='test')
faces, names, best_sim, _, _ = searcher.search(multi_face_img, 'test')

print(names)

result

>> reba

Match

Match the person between two pictures.

source, target, sim = searcher.match(face1, multi_face_img)
cv2.imshow("source", source[0])
cv2.imshow("matched", target[0])

Get embedding vectors

Get the raw embedding vectors of faces in given image.

# expect a tensor shape with (3, 512)
emb, boxes, landmarks = searcher.embedding_faces_in_the_wild(multi_face_img)