A convolutional network for face detection. This work is based on "Multi-view Face Detection Using Deep Convolutional Neural Networks" by Farfade et al., although some implementation details differ.
Following scripts are provided:
- scripts/download_data.py
- scripts/train_model.py
- scripts/visualization.py
- scripts/accuracy.py
Downloads Celeb Dataset and performs simple preprocessing. Please note that Celeb Dataset comes with its own license that you need to abide by to use it. With a little bit of effort you could adapt this code to work with a different dataset.
Trains the network.
Provides a few handy functions for visualizing data batching results, predictions for image crops, face detections and face heatmaps.
Provides functions to test accuracy of the trained network. OpenCV CascadeClassifier can also be tested for comparison.