Data set of the paper U-Can't (re)touch this – A Deep Learning Approach for Detecting Image Retouching from Daniela Aumayr and Pascal Schöttle for ICIAP2021.
https://doi.org/10.1007/978-3-031-06430-2_11
First, we used scikit-image (skimage) to crop the original images from RAISE to a square format, resize them to 256x256 pixel and then save them in JPEG format with quality factor 100. Since Snapseed is a mobile application, the Android emulator Bluestacks is used to edit the photos with Snapseed. In Snapseed, we processed the 1 000 images with nine different filters, creating a data set of 10 000 images in total.
The filters we used to create our retouched images (see Figure 1) are numbered as follows:
01 Original – original image
02 Smooth – blur
03 Pop – higher contrast
04 Accentuate – higher contrast and more intensive colours
05 Morning – bottom part of image has warm colouring
06 DramaDark – higher contrast, image will be darker
07 Vintage – Vintage Look, brown-filter
08 Retrolux8 – image gets a used look, like old pictures
09 Crunch3 – image gets a texture and dark vignette
10 Grain104 – grain is added to image