SciTextures

The SciTextures dataset contains over 100,000 images of textures and visual patterns generated by various scientific models and simulations, combined with the code that generated these images. The models cover a wide range of scientific and artistic domains from physics, chemistry, biology, tech, and art. The dataset is free to use under the CC0 license. 

SciTextures is focus on textures images. For PBR materials made by the similar approach see the Emergent Textures Dataset.

Gallery1     Gallery2    Gallery3      Gallery4

Download

Samples: JPG1, JPG2, JPG3, JPG4, zip1,  zip2,   zip3

Full Dataset JPG  format (100,000 images 12gb)

Sample Dataset (1200 images 174mb)

Full dataset PNG format  Part1,   Part2,   Part3,   Part4, Part5

Resources

Other download sources: Zenodo, HuggingFace

Code for all models/simulation in dataset

Emergent Textures dataset: similar approach but to PBR materials

Code structure

The code for each model is given in the datafolder.

The code structure of all mode is the same:

All the code is contain in the file: generate.py

The code can be run using the function:  

def generate_texture(outdir: str, sz: int = 512, num_samples: int = 20):

outdir: ouput folder where the images will be saved.

sz: size of each image in pixels (across single dimension)/

num_samples: number of different images to generate