I have a dataset that has 12 classes in the base directory. However, these 12 classes consist of several amounts of Images. The number of images of 12 classes is inconsistent therefore its impacts the total accuracy. Thus, should I apply the data augmentation to the particular classes that have a low amount of data?

Image data of each class:

#Dummy Classes

[AAAA: 713
ABCD: 274
ACBD: 335
ADBC: 576
BBBB: 538
BACD: 607
BCAD: 253
BDAD: 257
CCCC: 463
CABD: 309
CBAD: 452
CDAB: 762]

Therefore, if should I have to apply data augmentation to increase the amount of data in that particular classes that have a lower amount of image, as a result, I have been applied data augmentation but it does not increase the image data. Besides that, I want to generate the augmented data with the raw data that means the input and out directory will be the same. Therefore, it is possible to assist me to solve this problem?

Notebook: Google Colab OS: Windows 10

Code of Augmentation for particular (Individual Classes):

from keras.preprocessing.image import ImageDataGenerator

datagen = ImageDataGenerator(
    shear_range = 0.2,
    zoom_range = 0.2, 
    fill_mode = 'reflect', cval = 125)

i = 0

for batch in datagen.flow_from_directory(directory = ('/content/dataset/ABCD'),
                                         batch_size = 317,
                                         target_size = (256, 256),
                                         color_mode = ('rgb'),
                                         save_to_dir = ('/content/dataset/ABCD'),
                                         save_prefix = ('aug'),
                                         save_format = ('png')):
  i += 1
  if i > 100:

Output: Found 0 images belonging to 0 classes.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.