MTCNN face detection & alignment all in TensorFlow
Introduction
This is a demo for MTCNN implementation all in TensorFlow api to take advantage of GPU computing resource.For more details of MTCNN, please refer to the paper arXiv paper.
Dependencies
- TensorFlow 1.4.1
- TF-Slim
- Python 3.6
- Ubuntu 16.04
- Cuda 8.0
Usage
First you should run 'python npy2ckpt.py' to convert the three npy files(get from facenet) for pnet/rnet/onet to one checkpoint if you do not have the checkpoint file(Note:the three npy files and converted checkpoint file already in mtcnn_model of this repository).
Then replace your pictures in 'examples' and run 'python demo.py'.
Result
demo_result:
References
- Kaipeng Zhang, Zhanpeng Zhang, Zhifeng Li, Yu Qiao , " Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks," IEEE Signal Processing Letter(https://arxiv.org/pdf/1604.02878.pdf).
- facenet