python - Keras - flow_from_directory: How to get number of samples in each category?

Question

Is there a way in Keras how to get count of images within each category when using ImageDataGenerator? Let's say I have train folder with 3 subfolders (classes) and each containing some number of images.

imagedatagen = ImageDataGenerator(...)    

imageflow = imagedatagen.flow_from_directory(
  source_dir, 
  class_mode='categorical', ...)

I can get class mapping or number of all samples with following commands:

class_dict = imageflow.class_indices # returns ie. {'aligator': 0, 'cat': 1, 'dog': 2}
sample_count = imageflow.samples # returns sum of all samples, not divided into categories

But how may I get info, that category 0 contains 1000 pictures, category 1 contains 1500 pictures and category 2 all the rest? I know that I can iterate all files with help of os module, but would like to know if there is some other way and I could not find it.

Answer 1

As far as my knowledge there is no built-in approach to this.

You can try with collections module.

from collections import Counter

counter = Counter(imageflow.classes)

print(counter.items()) # dict_items([(0, 1648), (1, 3614)])