To run the codes, first pip install -r requirements.txt. Preferably in a virtual environment.
Then, download the vgg_face_matconvnet.tar.gz file from here. Download the face_gender dataset. Once the dataset is extracted, the faces folder should also contain the following 5 .txt files
- fold_0_data.txt
- fold_1_data.txt
- fold_2_data.txt
- fold_3_data.txt
- fold_4_data.txt
Add the paths to the config/config.json directory,
- "gender_data_path": <path to gender dataset downloaded earlier>
- "path_mat_file": <path to vgg descriptor model stored in vgg_face_matconvnet.tar.gz>
Other parameters can be changed here accordingly.
To run the classifier model run the following command (make sure you're in the directory that the main files are stored):
python main_gender.py -c ./configs/config.json
The project contains the following structure:
- Model directory contains the tensorflow model for VGG descriptor, as well as a model that uses this as a feature extractor to do gender classification.
- Trainer contains the training logic
- data_loader contains data pipeline
- Data folder to store data (not necessary)
- Utils are anything additional that doesn't fit in the other folders.