Describable Textures Dataset: Textural Images with Human-Centric Attributes

The Describable Textures Dataset (DTD) is an evolving collection of textural images annotated with human-centric perceptual attributes. It is made available to the computer vision community for research purposes by the Visual Geometry Group at the University of Oxford. The dataset was last updated on the Hugging Face platform on 2023-05-11.

Use Cases

• Train texture classification models based on human-centric perceptual attributes.
• Benchmark algorithms for material recognition in unconstrained environments.
• Research the relationship between visual texture and descriptive language.
• Develop models for zero-shot or few-shot texture recognition based on semantic attributes.

Strengths

• Dataset is described as an 'evolving collection', suggesting ongoing curation.
• Annotations are based on human-centric perceptual properties of textures.
• Source is the Visual Geometry Group at the University of Oxford, a known research institution.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown.

Provenance

Source

Visual Geometry Group, University of Oxford (robots.ox.ac.uk/~vgg/data/dtd/)

Collection Method

Collection of textural images in the wild, annotated with human-centric attributes.

Freshness

Last updated 2023-05-11 07:36:27; freshness should be verified.