mxnet-finding-tiny-face

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

Getting Started

• Download the origin author's model from here.
• Transform the origin Matconvnet model to MXNet model using matconvnet_hr101_to_mxnet.py.

Declaration:

• the code is used to do face detection

• forked by chinakook/hr101_mxnet

• and added some Batch processing codes to read images from a folder named"./data/images_stair_enter/"

• then save them in the result file "./results/test_stair_enter_results/"

• average processing time is 2.6 seconds

• total processing time is 164.96s of processing 55 images which is obviously much faster than the original code by matlab（https://github.com/peiyunh/tiny） which is 1060s.

Using method:

• python mx_tiny.py