Applying Mult-task Cascade Convolutional Neural Network (MTCNN) for Face Detection
About
Multi-task Cascade Convolutional Neural Network (MTCNN) Architecture.
This repository has as goal to use the Multi-task Cascade Convolutional Neural Network (MTCNN) for face detection. Basically, MTCNN is devided in three steges. The first stage contain a prediction of candidate facial regions (P-Net), the second stages filters the bounding boxes (R-net), and the third and last stage proposes facial landmarks (O-Net). For know more about the technique, please read the referenced paper (find in the section call Reference).
Datasets Used
Real and Fake Face Detection dataset, available in kaggle competition.
For apply the MTCNN approach was choosed 10 images. The first nine images is available in the dataset called Real and Fake Face Detection, this dataset can be acess free from the kaggle competetion. The last image was capture by photographer Bob N. Renee.
Getting Started
Prerequisites
Good knowledgment in Deep Convolutional Neural Network, Computer Vision, and know how to use ".sh" file.
Installing
First thing first, install the mtcnn package, that contain the implementation of architecture and packages dependecie:
pip install mtcnn
Usage
For run this repository, please open the terminal and run the following command:
sh run.sh
Result
Result of application Multi-task Cascade Convolutional Neural Network (MTCNN) architecture. Photo by Bob n Renee, some rights reserved.
When you run the run.sh file, you get as result the face and landmark detection of each image. The figure bellow show how the result of MTCNN techniques it is.
Reference
[ZHANG2016] Zhang, K., Zhang, Z., Li, Z., and Qiao, Y. (2016). Joint face detection and alignment using multitask cascaded convolutional networks. IEEE Signal Processing Letters, 23(10):1499–1503.
Sincerely: Neemias B. da Silva