• Description:

The Describable Textures Dataset (DTD) is an evolving collection of textural images in the wild, annotated with a series of human-centric attributes, inspired by the perceptual properties of textures. This data is made available to the computer vision community for research purposes.

The "label" of each example is its "key attribute" (see the official website). The official release of the dataset defines a 10-fold cross-validation partition. Our TRAIN/TEST/VALIDATION splits are those of the first fold.

Split Examples
'test' 1,880
'train' 1,880
'validation' 1,880
  • Feature structure:
    'file_name': Text(shape=(), dtype=string),
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=47),
  • Feature documentation:
Feature Class Shape Dtype Description
file_name Text string
image Image (None, None, 3) uint8
label ClassLabel int64


  • Citation:
Author    = {M. Cimpoi and S. Maji and I. Kokkinos and S. Mohamed and A. Vedaldi},
Title     = {Describing Textures in the Wild},
Booktitle = {Proceedings of the {IEEE} Conf. on Computer Vision and Pattern Recognition ({CVPR})},
Year      = {2014} }