This is the sample code for my article about fast feature extraction using Keras. You can find it at the following link.
https://duongnt.com/fast-feature-extraction
from image_to_hdf5 import write_data_to_hdf5
write_data_to_hdf5('dataset/train', 'hdf5_data/train.hdf5')
write_data_to_hdf5('dataset/valid', 'hdf5_data/valid.hdf5')
write_data_to_hdf5('dataset/test', 'hdf5_data/test.hdf5')
You can use the HDF5Generator class to yield data from hdf5 files.
from hdf5_generator import HDF5Generator
with HDF5Generator('C:/Learn/Python/fast_feature_extraction/hdf5_data/test.hdf5', 32) as hdf5_gen:
for images, labels in hdf5_gen.generator():
print(images.shape, labels.shape)
break
Will print
(32, 8192) (32,)
Warning: although training a single classifier does not take much time, we will perform hyperparameters tuning for 160 trials. On my low-end GPU, the whole process took around 90 minutes.
from fast_feature_extraction import start_hyper_tuning
start_hyper_tuning()
After tuning, the best model will be saved as gender_prediction_ffe_best.keras. Also, the test loss/test accuracy will also be printed to console (your loss/accuracy will be somewhat different because of the random initialization of the network weights).
[0.6207340359687805, 0.9470587968826294]
MIT License