This is an inference-only implementation for MXNet of tiny face. See the following references for more information:

"Finding Tiny Faces."
Peiyun Hu, Deva Ramanan
arXiv preprint arXiv:1612.04402, 2016.

https://arxiv.org/abs/1612.04402

• Download the origin author's model from here.
• Transform the origin Matconvnet model to MXNet model using matconvnet_hr101_to_mxnet.py.
• To run:
  • simply run tiny_detection_mxnet.py
  • use from your code

from tiny_fd import TinyFacesDetector
import cv2 as cv

detector = TinyFacesDetector(model_root='./', prob_thresh=0.5, gpu_idx=0)
img = cv.imread('./selfie.jpg')
boxes = detector.detect(img)

You should install opencv-python 2.4 to get the nearly completely identical result to the origin implemetation.

cydonia999's Tensorflow implemetation