Meta-Album Textures: 8,675 Preprocessed Texture Images for Manufacturing

8,675 images across 64 texture classes, preprocessed to 128x128 pixels. This dataset is a curated combination of four established texture datasets (KTH-TIPS, KTH-TIPS 2, Kylberg, and UIUC) collected in laboratory conditions. It was compiled and released by Ihsan Ullah for the Meta-Album project in March 2022.

Use Cases

• Training image classifiers for texture recognition based on the 64 material classes.
• Benchmarking few-shot or meta-learning algorithms on a standardized, preprocessed image dataset.
• Developing models for quality inspection in manufacturing by analyzing material surface textures.

Strengths

• Consists of 8,675 images, providing a substantial sample for model training.
• Images are uniformly preprocessed to 128x128 pixels, ensuring consistent input dimensions.
• Combines four established academic datasets (KTH-TIPS, Kylberg, UIUC), likely offering diverse texture samples.

Limitations

• Images were collected in controlled laboratory conditions, which may not reflect real-world variability.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is under a non-commercial license (CC BY-NC 4.0), restricting commercial use.

Provenance

Source

Combination of KTH-TIPS, KTH-TIPS 2, Kylberg Textures, and UIUC Textures datasets.

Collection Method

Images were captured in controlled laboratory environments, then cropped, squared, and resized to 128x128 pixels.

Time Range

Original source datasets were created between 2004 and 2010.