Meta-Album Textures (Mini): 64 Classes of Laboratory-Captured Images

2560 preprocessed texture images across 64 classes, sourced from four established academic datasets. The Meta-Album Textures (Mini) dataset was created by Ihsan Ullah in March 2022, combining and standardizing images from KTH-TIPS, Kylberg, and UIUC texture collections. All images are cropped to squares and resized to 128x128 pixels with an anti-aliasing filter.

Use Cases

• Training image classifiers for texture recognition based on 64 distinct material classes.
• Benchmarking few-shot or meta-learning algorithms on a standardized image dataset.
• Developing models for quality inspection in manufacturing using controlled laboratory image data.
• Studying the generalization of vision models across combined, preprocessed datasets from multiple sources.

Strengths

• Contains 2560 images, providing a defined scale for model training and evaluation.
• Images are preprocessed to a uniform 128x128 pixel size, ensuring consistency.
• Combines four established academic texture datasets (KTH-TIPS, KTH-TIPS 2, Kylberg, UIUC), offering a diverse benchmark.

Limitations

• Dataset size (2560 images) may be limited for training large, deep learning models from scratch.
• Images were captured in controlled laboratory conditions, which may not reflect real-world variability.
• Column-level documentation and sample data are unavailable, requiring inspection after download to understand data structure.

Provenance

Source

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

Collection Method

Images were captured in laboratory conditions with configurable brightness, luminosity, scale, and angle, then preprocessed (cropped and resized).

Time Range

Original source datasets were created between 2004 and 2010.