A Neural network trained YouTube Faces Dataset on to detect the facial keypoints from images. This is the first project of the Udacity Computer Vision Nanodegree.
This method uses Haar Cascades in images in order to locate RoIs (Region of interests), which are then given to a CNN to generate facial keypoints. The CNN architecture is pictured below:
For a more comprehensive understanding of this project, unzip the jupyter notebooks found in the Notebooks directory and iterate through them.
The dataset used can be found here
You can download the dataset from this link Put the contents in the data directory
The following libraries are mandatory: torch, torchvision, matplotlib, numpy, pandas, cv2, pillow
To train and test the whole pipeline, run the following command:
python network.py
The image below is an example of how this network isolates a face region and then applies the one pair of the predicted facial keypoints.
This model is very simple and should be treated as a quick solution to detecting points of interest on human faces.
Debucean Caius-Ioan @Udacity