ImageDataGenerator

keras ImageDataGenerator for Multi Labeled Data

Keras ImageDataGenerator is useful API for loading train/val/test dataset from data array or file system it also provide data augmentation method.

However unfortunatly Keras ImageDataGenerator not yet support Multi Labeled Dataset.

meaning of Multi Labeled Dataset is that one data has multiple labels.

for example, above image data have two labels which was fire and smoke.

for supporting Multi Labeled Dataset on Image Data Generator, i modified some part of keras.preprocessing.image

How to use?

import utils to your python script

from utils import *

this utils.py file almost same with keras.preprocessing.image without some function

make data file structure like bellow(Fire-Smoke folder have multi labeled images)

├── Fire(folder)
│   └── XXX-1.jpeg
│   └── ...
├── Smoke(folder)
│   └── XXX-1.jpeg
│   └── ...
├── Fire-Smoke(folder)
│   └── XXX-1.jpeg
│   └── ...
├── Bg(folder) # if you want
│   └── XXX-1.jpeg
│   └── ...

if you add 'Bg' folder, ImageDataGenerator(with multi-categorical) consider this folder with class 0

then, setting parameter 'class_mode' to "multi_categorical" in function "flow_from_directory"

if subdirectory have some seperator("-"), this utils will automatically identify whether this subdirectory is multi labeled class or not

Test ImageDataGenerator for Multi Labeled Dataset

first modify igd_test.py

if __name__ == '__main__':
    data_dir = 'path/to/dataset' # modify to your dataset path
    train_dir = os.path.join(os.path.abspath(data_dir), 'train')  # Inside, each class should have it's own folder
    validation_dir = os.path.join(os.path.abspath(data_dir), 'val')  # each class should have it's own folder

    get_data(train_dir, validation_dir)

    # release memory
    k.clear_session()

and running below command

python idg_test.py

then, some number of train/val data and labels are printed on your terminals

Reference

[1] keras.io
[2] https://github.com/keras-team/keras-preprocessing