Transmute2K: Library of Image Transformations Simulations and Augmentations
Transmute2K: Library of Image Transformations Simulations and Augmentations, based 2K methods, models and simulation from science, art and math. Transmute2K is a large-scale library and dataset containing 2,000 image transformation methods based on techniques, simulations
Transmute2K: Library of Image Transformations Simulations and Augmentations, based 2K methods, models and simulation from science, art and math.
Transmute2K is a large-scale library and dataset containing 2,000 image transformation methods based on techniques, simulations and models from various fields of science, math and art that have been adapted for image manipulation and image-processing.
In general, the transformations can be applied to standard RGB images, as well as to images or maps of other types, such as PBR materials and hyperspectral maps.
Each transformation follows a model or function. This can range from simple operations such as rotation and blurring to more complex physical and scientific models involving diffusion, clustering, topology,camera effect, optica, physical and chemical proccess and other forms of simulation, with the image used as the working medium on which the transformation operates on.
The transformations are divided into two groups:
1. Image-to-image transformations, which transform one image into another according to a specific model, such as cross-fading.
2. Single-image transformations, which modify one image using a specific process, such as blurring or diffusion.
Each transformation function receives either one image or a pair of images and outputs a sequence of images corresponding to different stages of the transformation.
The input can be a standard RGB image or another type of map, such as a PBR material map or a spectral map.
The transformations were collected from a wide range of fields. Some use standard image-augmentation or image-modification functions, while others adapt models and simulations from different scientific and artistic domains and apply them to images.
The dataset was created/collected using an agentic AI pipeline based on several large language models, (GLM, KIMI) with human manual inspection and filtering of the results.
File Structure:
Code Files:
Python scripts for all approximately 2,000 transformations are available in:
Code_image2image_transformations.zip
Code_single_image_transformations.zip
The dataset-generation pipeline is available in:
dataset_generation_code.zip
Image Transformation Sequences
Sequences of images generated by each transformation are available in files whose names begin with **transformation_sequence**
Each file contains an sequence of images of every transformation in the dataset applied to a random image or pair of images.
Image Transformation As Animated GIFs:
Files whose names begin with **GIF** contain examples of each transformation as animated GIFs.
PB📤 Share this page
Found this useful? Share it with your network.
📤 Share this page
Found this useful? Share it with your network.
Files are hosted on the source repository. Click download to access the full dataset.