cnn-facefeature-detection

Overview

This is a quick 4-layers single pool CNN for detecting 15 face keypoints.

Dataset

The dataset is using Montreal dataset. It can be downloaded here

Usage

1. Git clone
2. Install all dependancies:

• Python 3.5 (or above)
• Numpy
• TensorFlow (1.13.x above)
• Matplotlib
• Pandas

3. Download the dataset
4. Configure test.py, adjust path
5. Build the model. Visualization is provided.

Accuracy

Accuracy is rather low, lot of missing facial keypoints in the training data. Might need better dataset.