Texture and Language
The goal of this project is to capture characteristics of textures in images using natural language. We have collected a Texture/Language dataset based off of the Describable Texture Dataset (Cimpoi et al, 2014), where AMT workers describe texture images in their own words, including attributes such as color, pattern, and style in their descriptions. We aim to use this dataset for classifying attributes of texture-rich images, generating natural language descriptions of textures, and performing image retrieval from natural language.
(Update 6/19): We presented a poster in the 6th Fine-Grained Visual Categorization Workshop at CVPR 2019 in Long Beach, on Visualizing and Describing Fine-grained Categories as Textures, using our Texture/Language Dataset. View the project page and poster here! Read the paper here!
(Update 11/18): We recently presented our progress at Harvard University in the New England Computer Vision Workshop! You can preview the output of our first round of data collection here!
Thanks for checking in! - Mikayla