Age-BMI-Gender Estimation

This is a Tensorflow-Keras implementation of a CNN for estimating visual-BMI, age and gender from a face image. In training, [the IMDB-WIKI dataset, especially wiki_crop version dataset is for age-gender model training, while the BMI dataset is for visual-BMI model training].

###[Log of modification history]

• [Sept. 20, 2020] get to consider and do some research on this topic
• [Oct. 7, 2020] Ready to do the coding Taking Another Tensorflow-based project as reference

Dependencies

• Python3.6+， Tensorfow v1.14+,
• Other modules in python: -- Dlib / MTCNN for running the demo.py

DEMO Tested on:

• MacOS10.15.5 Catalina, Python 3.6.9, Tensorflow 1.14

Usage of DEMO

Use trained model for demo

• Run the demo script (requires web cam). You should run the DEMO in the sub-directory of the project path,
  " test_ABG_estimation ", You can use --model_func [function_name] , these arguments indicating the model-function to demo ! "(bmi-face, age-face, gender-face, abg-face), exclusively. --image_dir [IMAGE_DIR] , optionally to use images in the directory instead. Othersie the WebCam is employed as default. This directory is absolute path. --weight_file [MODEL located_DIR], path to weight file (e.g. *** weights.***.h5)

python demo.py --weight_file trained_model --model_func bmi-face --image_dir /Users/YourName/Pictures/test-ABG-model

Training Age-BMI-Gender model

Create BMI training data from the BMI-Skymind dataset

Firstly, manually download the raw dataset from the url provided:

Secondly, using the following script to generate transferred crop-faces dataset version

python create_db_bmiface_with_margin.py --input original_bmi_dataset --output test_abg_estimation/bmi_face_data

Thirdly, using the following script to generate the dataset label auxilary file in MATLAB format, *.mat.

python creare_db_bmi.py --input bmi_face_data --output bmi_face_data --img_size 224 

The resulting files with default parameters are included in this repo (meta/imdb.csv and meta/wiki.csv), thus there is no need to run this by yourself.

Train BMI model for the Visual-Face BMI function

Train the model architecture using the training data created above:

python train_bmi_model.py --bmi_dataset bmi_face_data

Trained weight files are stored as checkpoints/*.h5