benckx/iapetus-images

Images manipulation tools (resize, crop, compress, rotate, reframe around faces, etc.) written Java and Kotlin to clean up images datasets (e.g. for machine learning).

About

Collection of functions to prepare datasets for Machine Learning training (resize, crop, cut pieces, create mosaic, etc.)

ImagesUtils

Various functions to resize, compress, rotate, etc. individual images (in Java).

ImagesFolderUtils

Methods that apply operation on an input folder (and its sub-folders recursively) containing images (in Kotlin). It doesn't override the original images, but create an output folder with the processed images.

List all images in folders and sub-folders:

File.listAllImages()

Crop all images found in input folder to proportions (e.g. 1x1, 16x9, etc.):

File.cropImagesToProportions(x = 1, y = 2)

Resize all images found in input folder:

File.resizeImagesTo(width = 1920, height = 1080)
File.resizeImagesTo(size = 1024)

Create Mosaics

Create e.g. 4K images from 1080p images.

File.createMosaics(width = 3840, height = 2160, nbrOfOutputImages = 2000)

Chain calls

All those functions return a File, and therefore can be chained:

val folder = File("/home/benoit/Pictures/dataset/my_data_set")

folder
    .cropImagesToSquares()
    .resizeImagesTo(1024)
    .randomizeImageNames()

drwxrwxr-x  2 benoit benoit 3,8M Oct 13 18:05 my_data_set
drwxrwxr-x  2 benoit benoit 3,9M Oct 13 18:12 my_data_set-square
drwxrwxr-x  2 benoit benoit 3,8M Oct 13 18:18 my_data_set-square-1024
drwxrwxr-x  2 benoit benoit 2,2M Oct 13 18:18 my_data_set-square-1024-rnd

FaceUtils

Crop pictures around the faces (in such a way that faces are always at the same place in the output images), based on information contained in a CSV file. You can build such a CSV with this project: https://github.com/benckx/tensorflow-face-detection

FaceUtils.reFrameAroundFace(csv: String, frame: ImageContainingFace, output: String)

For example, with this input image and frame (frames outlines added for visualization purpose; the CSV face data is in blue, the WAIST_LEVEL frame in red and orange):

val WAIST_LEVEL = ImageContainingFace(640, 1080, Face(Rectangle(170, 170, 300, 300)))

It will extract the following:

Import with Gradle

repositories {
    maven { url "https://jitpack.io" }
}

dependencies {
    compile "com.github.benckx:iapetus-images:master-SNAPSHOT"
}