how can I train network with mnist one class only?

[lerndeep]

No description provided.

[lerndeep]

I also want to generate one class 1000 images only please let me know.

[hcnoh]

Hello! Thank you for visiting my repo. This repo does not provide currently the situation that you want. But I can help you to modify the code so that you can train only one class.

If you check the train.py line 27 to 29, you can find the following part:

features = np.vstack([loader.train_features, loader.test_features]).astype(np.float32)

num_sets = loader.num_train_sets + loader.num_test_sets

This part is what you have to modify. The following modification can be helpful to you:

features = np.vstack([loader.train_features[np.where(loader.train_labels == 0)], loader.test_features[np.where(loader.test_labels == 0)]]).astype(np.float32)

num_sets = loader.train_features[np.where(loader.train_labels == 0)].shape[0] + loader.test_features[np.where(loader.test_labels == 0)].shape[0]

This example is for only using class 0, about 7000 images (training 6000 + test 1000).