jackhuntcn / facial_beauty_prediction

facial_beauty_prediction

A Facial Beauty Prediction web demo, using dataset from SCUT-FBP5500-Database. Inspire by https://towardsdatascience.com/how-attractive-are-you-in-the-eyes-of-deep-neural-network-3d71c0755ccc MSE loss on test data is 0.16.

How to run this demo

git clone https://github.com/jackhuntcn/facial_beauty_prediction.git
cd facial_beauty_prediction/web/
# download model.h5 from BaiDuYun and put it into model/ directory, make sure model/model.h5 is exists
# https://pan.baidu.com/s/1oJd67FMwO9KVwN5hgda3JQ password: mw8m
python app.py

requirements

• numpy
• scipy
• cv2
• keras 2.1.5
• flask / flask-uploads
• gevent

update

using generator to reduce memory usage, see FBP5500-Resnet50-fit_generator.ipynb and using a much small lr to get a pretty good mse loss (0.0885)