How to allow for preprocessing of image channel values in ImageDataGenerator and flow_from_directory?

[pchankh]

In short we only see rescale option in ImageDataGenerator. How do you do the following for pertained VGG16 network that needs the channels values preprocessed as follows:
im = cv2.resize(cv2.imread('cat.jpg'), (224, 224)).astype(np.float32)
im[:,:,0] -= 103.939
im[:,:,1] -= 116.779
im[:,:,2] -= 123.68
im = im.transpose((2,0,1))
im = np.expand_dims(im, axis=0)
Where do we specify these options in ImageDataGenerator?

Thanks
Dr

[mabirck]

There is a tread #3338 and gives a solution by adding a pipeline with fork of imageGenerator in keras. I think is that your looking for, sorry if I am wrong.

A more manual way of doing this is:

Pass the batches created in the form below trough a method that treats you data as you want it.

for e in range(nb_epoch):
    print 'Epoch', e
    batches = 0
    for X_batch, Y_batch in datagen.flow(X_train, Y_train, batch_size=32):
        loss = model.train(X_batch, Y_batch)
        batches += 1
        if batches >= len(X_train) / 32:
            # we need to break the loop by hand because
            # the generator loops indefinitely
            break

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs, but feel free to re-open it if needed.