Implementation of a CNN. The architecture does not include training modules. The weights should be pre-trained.
Weights are extracted from a json file of pre-trained weight matrices.
The weight dictionary is of the form
Wl i j - for convolutional layer weights, where
l - overall layer number i - ith feature in the next layer j - jth feature in the previous layer and
Wl for mlp weights
l - overall layer number
For demo
- face_test.py - contains a gender classification demo on webcam by faces detected by opencv's face detection
- face_det.py - gender classification demo on webcam by face detection tracking using Optical Flow
- test.py - contains gender detection demo on still images For usage, see the test.py file