Age and Gender estimation

Age and gender estimation using CNN. The model was trained on the IMDB and Wiki dataset. The APPA-real-realease dataset was used for test phase.

Requirements

• `Tensorflow >= 2.0.1
• mtcnn
• opencv

See requirements.txt for full requirements. All packages can be installed by pip easily.

Train the model

At first, download face-only datasets from the dataset home page, extract and put them anywhere you'd like. The project structure should be as follows:

-- age-gender-estimation
 |
 |-- data
 |    |
 |    |-- imdb_crop
 |    |-- wiki_crop
 |    |-- appa-real-release
 |
 |-- train.py
 |-- README.md
 ...

Train the model with the following command.

python train.py --batch_size 128 --epochs 100 --lr 1e-3 --data_dir data --save_dir saved_models

When the training process finished, a file named history.npy should be saved to folder saved_models as default. Run python plot_curve.py --input_path saved_models/history.npy for visualization.

Test the model

We also provided a pretrained model to help you instantly test it. If you'd like to test on a single image, run python test_on_image.py --model_path MODEL_PATH --image_path IMAGE_PATH. Otherwise, in case you'd like to test on realtime camera, run python test_on_camera.py --model_path MODEL_PATH --image_path IMAGE_PATH.

Contact

Any question could be left as issues. Contact me via email [email protected]. You're all welcome.