| --- |
| license: cc0-1.0 |
| task_categories: |
| - image-segmentation |
| - zero-shot-classification |
| tags: |
| - textures |
| - PBR materials |
| size_categories: |
| - 100M<n<1B |
| --- |
| # General |
| VasTexture is a large-scale dataset of textures and PBR materials extracted from real-world images. |
| The repository contains 500,000 highly diverse texture images and PBR materials. All assets are free to download and use for any purpose (CC0 license). |
| The dataset is divided into textures images, and PBR materials. Where texture image are simply crop of regions in images with uniform textures. |
|
|
|
|
| The PBR materials and textures were extracted from natural images using an unsupervised statistical approach (no human intervention). |
| As a result, the textures and PBR materials are significantly more diverse but less refined compared to assets made using manual and AI approaches. This dataset is more suitable for tasks needing a large number of highly diverse assets like building datasets or large scale procedural generation. |
|
|
| [##Project Website](https://sites.google.com/view/infinitexture/home) |
|
|
|
|
|
|
| ## File Structure |
| The dataset is composed into two assets types textures images and PBR materials. |
|
|
|
|
| Texture image files contain the world **Texture** in the file. |
| PBR materiasl files contain the world **PBR** in the file. |
| If the PBR are seamless/tilable, the world **seamless** will appear in the file name (note that for textures images this mean some modification was done on texture edges). |
|
|
| If the PBR is 512x512 or larger, the world **large** will appear in the file name. |
| Most files will have texture size in their name. |
|
|
| Each file contain between 1,000 to 40,000 assets. |
| Files with the the word **sample** contain few dozens to few hundered samples. |
|
|
| ## Data generation code: |
|
|
| The Python scripts used to extract these assets are supplied at: |
|
|
| Texture_And_Material_ExtractionCode_And_Documentation.zip |
| |
| The code could be run in any folder of random images extract regions with uniform textures and turn these into PBR materials. |
| |
| Code for transforming data to seamless available at [https://github.com/sagieppel/convert-image-into-seamless-tileable-texture](https://github.com/sagieppel/convert-image-into-seamless-tileable-texture) |
| |
| |
| # GITHUB and Alternative download sources: |
| GitHub: [Texture/PBR extraction](https://github.com/sagieppel/Unsupervised-extraction-of-textures-and-PBR-materials-from-images), [Texture To Seamless](https://github.com/sagieppel/convert-image-into-seamless-tileable-texture) |
| |
| https://sites.google.com/view/infinitexture/home |
| |
| https://zenodo.org/records/12629301 |
| |
| ## Papers |
| Main paper: |
| [Infusing Synthetic Data with Real-World Patterns for |
| Zero-Shot Material State Segmentation](https://proceedings.neurips.cc/paper_files/paper/2024/file/6ef4a4b387a5a547ea699f3df7fc1248-Paper-Datasets_and_Benchmarks_Track.pdf) |
| |
| More detailed: |
| [Vastextures: Vast repository of textures and PBR materials |
| extracted from real-world images using unsupervised methods](https://arxiv.org/pdf/2406.17146) |
| |
