Face detection using TinyFace

This package comes with a TinyFace face detector. The Original Model is ResNet101 from tinyface. Please check for more details on TinyFace. The model is converted into MxNet Interface and the code used to implement the model are from hr101_mxnet.


See below for an example on how to use bob.ip.facedetect.tinyface.TinyFacesDetector:

 1import matplotlib.pyplot as plt
 2from bob.io.base import load
 3from bob.io.base.test_utils import datafile
 4from bob.io.image import imshow
 5from bob.ip.facedetect.tinyface import TinyFacesDetector
 6from matplotlib.patches import Rectangle
 8# load colored test image
 9color_image = load(datafile("test_image_multi_face.png", "bob.ip.facedetect"))
10is_mxnet_available = True
12    import mxnet
13except Exception:
14    is_mxnet_available = False
16if not is_mxnet_available:
17    imshow(color_image)
20    # detect all faces
21    detector = TinyFacesDetector()
22    detections = detector.detect(color_image)
24    imshow(color_image)
25    plt.axis("off")
27    for annotations in detections:
28        topleft = annotations["topleft"]
29        bottomright = annotations["bottomright"]
30        size = bottomright[0] - topleft[0], bottomright[1] - topleft[1]
31        # draw bounding boxes
32        plt.gca().add_patch(
33            Rectangle(
34                topleft[::-1],
35                size[1],
36                size[0],
37                edgecolor="b",
38                facecolor="none",
39                linewidth=2,
40            )
41        )

This face detector can be used for detecting single or multiple faces. If there are more than one face, the first entry of the returned annotation supposed to be the largest face in the image.

Multi-Face Detection results using TinyFace.

Fig. 1 Multiple faces are detected by TinyFace.