hawk-eyes 2.0.8

Face recognize, Object tracking, OCR

Hawk-eyes

A library for Computer Vision, flexible and easy to use. The library uses a lot of models like ArcFace, RetinaFace, FaceLandmark, OCRModels .... These models will be downloaded automatically when used for the first time, please make sure your network connection is not blocked to google drive.

Install

Hawk-eyes is available on pypi.org, if you just want to use it for your project, install it using pip. Requires:

• python>=3.8
• torch==1.8
• onnxruntime==1.7 (or onnxruntime-gpu)

pip install hawk-eyes

Methods

Supported methods:

• Retina Face
• ArcFace
• Face similar
• face landmarks
• face angles
• Tracking
• Yolov5 inference

1. Face Recognition

1.1 Face detection

Retina Face: Single-stage Dense Face Localisation in the Wild

import cv2
from hawk_eyes.face import RetinaFace

retina = RetinaFace(model_name='retina_s')

img = cv2.imread('path/to/image.png')
bboxes, kpss = retina.detect(img)
for box in bboxes:
    box = box[:4].astype(int)
    cv2.rectangle(img, (box[0],box[1]), (box[2],box[3]), (0,255,0),thickness=2)
cv2.imshow('image', img)
cv2.waitKey(0)

1.2 Face extract feature use ArcFace

ArcFace: Additive Angular Margin Loss for Deep Face Recognition

import cv2
from hawk_eyes.face import RetinaFace
from hawk_eyes.face import ArcFace

arc = ArcFace(model_name='arcface_s')
retina = RetinaFace(model_name='retina_s')

img = cv2.imread('path/to/image.png')
bboxes, kpss = retina.detect(img)
for box,kps in zip(bboxes, kpss):
    emb = arc.get(img, kps)
    print(emb)

1.3 Get similar

Coming soon

1.4 Get face landmarks, face angles

import cv2
from hawk_eyes.face import RetinaFace, Landmark

retina = RetinaFace(model_name='retina_s')
landmark = Landmark()

img = cv2.imread('path/to/image.png')
bboxes, kpss = retina.detect(img)
for box,kps in zip(bboxes, kpss):
    land = landmark.get(img, box)
    angle = landmark.get_face_angle(img, land)
    print(angle)

2. Tracking

import cv2
from hawk_eyes.face import RetinaFace
from hawk_eyes.tracking import BYTETracker

bt = BYTETracker()
retina = RetinaFace(model_name='retina_s')

cap = cv2.videoCapture(0)
ret, _ = cap.read()
while ret:
    ret, frame = cap.read()
    boxes,_ = retina.detect(frame)
    boxes, ids = bt.predict(frame, boxes)
    print(ids)