image-preprocessing

Image pre-processing pipeline

Created at June 20, 2017

Korea University, Data-Mining & Information Systems Lab

Bumsoo Kim (meliketoy@gmail.com)

Requirements

• python 2.7
• OpenCV

sudo pip install opencv-python==3.2.0.8

Input directory

The input directory should be in the given format:

[:folder]
    |-[:class 0]
        |-[:img 0]
        |-[:img 1]
        |-[:img 2]
        ...
    |-[:class 1]
    |-[:class 2]
    ...
        ...
            ...

Modules

1. print

sudo python main.py print

This module will print all the the file names of image related file formats(".jpg", ".png")

2. read

sudo python main.py read original

This module will read all the images and print out the spacial dimension of image related files.

3. resize

sudo python main.py resize [:len]

# Example, to consist 256x256 images
sudo python main.py resize 256

This module will save all the resized images into your given directory

4. split

sudo python main.py split

This module will organize your input file directory into the following format. You should manually set how much validation sets you want in your val class in val_num from config.py.

[:folder]
    |-train
        |-[:class 0]
            |-[:img 0]
            |-[:img 1]
            |-[:img 2]
            ...
        |-[:class 1]
        |-[:class 2]
        ...
            ...
                ...
    |-val
        |-[:class 0]
            |-[:img 0]
            |-[:img 1]
            |-[:img 2]
            ...
        |-[:class 1]
        |-[:class 2]
        ...
            ...
                ...

5. check

sudo python main.py check

This will check how your data directory is consisted. An example for the file directory after running module 4 (split) is as below.

$ sudo python main.py check

| train set : 
    | false-positive   3345
    | true-positive    2547
| val set : 
    | false-positive    100
    | true-positive     100

6. augmentation

sudo python main.py aug

This module will apply various image augmentations and enlarge your training set. The input should be the splitted directory after running module 4 (split)