Pytorch_Retinaface is a GitHub repository that implements the RetinaFace face detection algorithm using PyTorch. It provides a robust and efficient face detection solution that can handle various face scales, poses, and occlusions in images and videos.

Pros

• High accuracy in face detection, even for challenging scenarios
• Supports both CPU and GPU inference
• Includes pre-trained models for easy deployment
• Provides options for different backbones (ResNet50, MobileNet) to balance speed and accuracy

Cons

• Limited documentation and examples for advanced usage
• Requires some understanding of PyTorch and deep learning concepts
• May be computationally intensive for real-time applications on low-end devices
• Lacks built-in face recognition capabilities (only performs detection)

Code Examples

1. Loading a pre-trained model:

from models.retinaface import RetinaFace
from data import cfg_mnet

# Load the model
net = RetinaFace(cfg=cfg_mnet, phase='test')
net.load_state_dict(torch.load('weights/mobilenet0.25_Final.pth', map_location=torch.device('cpu')))
net.eval()

2. Performing face detection on an image:

from utils.box_utils import decode, decode_landm
from utils.nms.py_cpu_nms import py_cpu_nms

# Preprocess the image
img = cv2.imread('path/to/image.jpg')
img = np.float32(img)
im_height, im_width = img.shape[:2]
scale = torch.Tensor([img.shape[1], img.shape[0], img.shape[1], img.shape[0]])
img -= (104, 117, 123)
img = img.transpose(2, 0, 1)
img = torch.from_numpy(img).unsqueeze(0)

# Perform detection
loc, conf, landms = net(img)
priorbox = PriorBox(cfg_mnet, image_size=(im_height, im_width))
priors = priorbox.forward()
boxes = decode(loc.data.squeeze(0), priors.data, cfg_mnet['variance'])
boxes = boxes * scale
boxes = boxes.cpu().numpy()
scores = conf.squeeze(0).data.cpu().numpy()[:, 1]

# Apply NMS
keep = py_cpu_nms(boxes, scores, 0.4)
dets = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=False)
dets = dets[keep, :]

3. Visualizing the results:

for b in dets:
    if b[4] < 0.6:
        continue
    text = "{:.4f}".format(b[4])
    b = list(map(int, b))
    cv2.rectangle(img, (b[0], b[1]), (b[2], b[3]), (0, 0, 255), 2)
    cx = b[0]
    cy = b[1] + 12
    cv2.putText(img, text, (cx, cy),
                cv2.FONT_HERSHEY_DUPLEX, 0.5, (255, 255, 255))

cv2.imshow("image", img)
cv2.waitKey(0)

Getting Started

1. Clone the repository:

   git clone https://github.com/biubug6/Pytorch_Retinaface.git
   cd Pytorch_Retinaface
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Download pre-trained weights:

   mkdir weights
   wget -P weights/ https://github.com/biubug6/Pytorch_Retinaface/releases/download/v1.0/mobilenet0.25_Final.pth
   

4. Run the demo:

   python detect.py