Textures-DTD: 5,640 Describable Texture Images for Computer Vision

Meta_Album_TEX_DTD_Extended is a preprocessed version of the Describable Textures Dataset (DTD) from the University of Oxford. It contains 5,640 images across 47 texture categories, sourced from Google and Flickr and annotated via Amazon Mechanical Turk. The Meta-Album version, created by Ihsan Ullah in March 2022, crops and resizes all images to 128x128 pixels.

Use Cases

• Training texture recognition models based on 47 describable texture categories.
• Benchmarking few-shot image classification algorithms using a standardized, preprocessed image set.
• Researching visual texture properties and descriptors in manufacturing and material science contexts.

Strengths

• Contains 5,640 images, providing a substantial base for model training.
• Offers 47 distinct texture class labels for multi-class classification tasks.
• Images are consistently preprocessed to a uniform 128x128 pixel size.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The original data collection process may introduce source bias from Google and Flickr.

Provenance

Source

Describable Textures Dataset (DTD), University of Oxford

Collection Method

Images collected from Google and Flickr, annotated via Amazon Mechanical Turk, and preprocessed by Meta-Album.