face_classification by Paddle

model	acc	acc(offcial repo)	acc(paper)
MiniXception	95.58% (MiniXception/best_model)	95% (gender_mini_XCEPTION.21-0.95.hdf5)	95%
SimpleCNN	95.76% (SimpleCNN/best_model)	94.88% (simple_CNN.81-0.96.hdf5)	96%

dataset

download imdb dataset (gender classification) from link and unzip it to dataset folder

tar -xf imdb_crop.tar

train

set images_path in line 33 of train.py and use follow script to start training

python3 train.py

eval

set images_path and 'model_path' in line 28 and 29 of eval.py and use follow script to start eval

python3 eval.py

demo

set gender_model_path in line 64 of demo.py and use follow script to start eval

python3 demo.py

sample