Meta-Album Textures: 800 Images of 20 Texture Classes for Manufacturing

Meta_Album_TEX_Micro is a preprocessed image dataset containing 800 texture images across 20 classes, sourced from the KTH-TIPS, Kylberg, and UIUC texture datasets. The dataset was created by Ihsan Ullah in March 2022 as part of the Meta-Album benchmark. All images are cropped to squares and resized to 128x128 pixels with an anti-aliasing filter.

Use Cases

• Train image classifiers to recognize 20 distinct texture types based on the described classes.
• Benchmark few-shot or meta-learning algorithms on a standardized, preprocessed texture dataset.
• Evaluate model performance on controlled, laboratory-captured images where variables like brightness and angle were configurable.

Strengths

• Contains 800 images across 20 distinct texture classes, providing a clear classification task.
• Images are preprocessed to a uniform 128x128 pixel size, ensuring consistency for model input.
• Combines data from four established academic texture datasets (KTH-TIPS, KTH-TIPS 2, Kylberg, UIUC).

Limitations

• Dataset size is relatively small with 800 images, which may limit training deep learning models from scratch.
• Images were captured in controlled laboratory conditions, which may not reflect real-world texture variability.
• Column-level documentation and sample data are unavailable, making detailed data inspection difficult prior to download.

Provenance

Source

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

Collection Method

Images were captured in laboratory conditions with configurable settings, then cropped and resized to 128x128 pixels.

Time Range

Original source data collected between 2004 and 2010.