Skip to content

Classification task hack to quickly learn vgg on cpu

Notifications You must be signed in to change notification settings

ashd97/bottleneck_vgg

Repository files navigation

Classification task for food, other, people.

We just take a pre-made network from top (which recognizes many different things), cut off several top layers, then putting own classifier there. And then quickly learn it on a small dataset keras-team/keras#4465 ImageNet itself has categories like food and person

First, we run the vgg model with a cut-off top on the data, and save the output to a file We will get generated data bottleneck_features_train.npy and bottleneck_classes_train.npy (wich is done by vgg_train_save_outputs.py). From these do the training "data" for our little model bottleneck_vgg_model.h5 Model learned with vgg_train_on_saved.py Then lets try to evaluate it using vgg_evaluate_on_saved.py with small dataset

Useful links

Especially thanks to hcl14

About

Classification task hack to quickly learn vgg on cpu

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages