Age Recognition Project

This project is for the development of a convolutional neural network (CNN) for assessing the age and gender of people's faces.

It forms part of Maggie Liuzzi's & Mitchell Clarke's Spring 2018 Research Project at UTS, and is developed in collaboration with the UTS Magic Lab.

The datasets explored are the following: a) Adience (https://www.kaggle.com/ttungl/adience-benchmark-gender-and-age-classification - 1GB), b) Wiki half of IMDB-WIKI (https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/ - In particular, "Wiki -> Download faces only (1GB)")

Installation

The project may be moved to a docker file at a later date, but for the time being, here is how you install it:

1. Install Python 2.7 if you haven't - (also works with Python 3.6). Also make sure that you have installed virtualenv. If you haven't, you can install it globally using pip:

pip install virtualenv

2. Clone this repo to a directory of your choice:

git clone https://gitlab.com/maggieliuzzi/agerecognition.git

3. Enter the folder and create a new virtual environment:

cd AgeRecognition
virtualenv -p [path to your python 2.7 interpreter] venv

4. Activate the new virtual environment (omit "source" if on Windows):

source venv/bin/activate

5. Once your environment is activated, install the project's dependencies:

pip install -r requirements.txt

With that, you should be good to go!

Alternatively, you can follow installation up to the end of Step 2 and then use PyCharm to create the virtual environment for you. If you do that, make sure pip is installed at /AgeRecognition/venv/bin and that the terminal in PyCharm says "(venv)" before the current location. If it does, you should be good to continue from the start of Step 5.

• Adience_Age_5y/ contains the required python scripts to train an age-recognition model with equi-width 5-year bins.

• Adience_Age_10y/ contains the required python scripts to train an age-recognition model with equi-width 10-year bins.

• Adience_Age_15y/ contains the required python scripts to train an age-recognition model with equi-width 15-year bins.

• Adience_Age_Equi_Depth/ contains the required python scripts to train an age-recognition model with equi-depth bins.

• Adience_Age_Equi_Depth/10_ED_bins/ contains the required python scripts to train an age-recognition model with 10 equi-depth bins.

• Adience_Gender/ contains the required python scripts to train a gender-recognition model.

• form...py scripts perform data processing to get the image list to a usable format.

• proc...py scripts separate data into training, validation and testing sets.

• train...py scripts train the network over a certain number of epochs and outputs an .h5 model.

• server...py scripts start a server that receives HTTP POST request with individual test images and outputs the estimated probabilities. Eg: http://0.0.0.0:4000/predict "prediction_age": { "1-15": 0.01101667433977127, "16-30": 0.9543766379356384, "31-45": 0.03458355739712715, "46-60": 0.000023107986635295674 }, "prediction_gender": { "Female": 0.10200900584459305, "Male": 0.8979910016059875 }

• test...py scripts test the quality of a model with the images in the test/ folder generated running proc...py.

• predict_gender_age.py takes a gender-recognition model, an age-recognition model and an image as arguments and predicts the gender and age of the person in the image.