pytorch-imagefolder

Create an Image Folder dataset layout from a CSV file with labeled filenames and classes (e.g. from Kaggle Data Science Bowl 2015 Plankton)

What it does

If you have a labeled csv file like the following:

• file_name,class
• image01.jpg,0
• image02.jpg,1
• image03.jpg,2
• etc..

Then this script can create the following structure:

data/
  - train/
      - class_1 folder/
          - img1.png
          - img2.png
      - class_2 folder/
      .....
      - class_n folder/
  - val/
      - class_1 folder/
      - class_2 folder/
      ......
      - class_n folder/

How to use

The following example might give you an overview on how to use this script or change it to work for your needs.

• csv file named labels.csv with column 1 being file names and column 2 being class names
• dataset that is used for training, use train
• image files in folder train_images
• are currently in directory of labels.csv

You would use the following execution of our script: /ImageFolderPyTorch.sh train_onelabel.csv train train_images

Where is this useful?

I used it for a Applied Machine Learning course where we had to do image classification for this kaggle project and PyTorch's ImageFolder Dataset class structure.