Many researchers in psychology, linguistics, neuroscience, and computer science are interested in using object concepts or object images in their studies. But selection of concepts and images often happens in an "ad hoc" fashion. This can lead to selection bias in the results. In addition, finding high quality images of objects useful for experiments is often cumbersome. To overcome these issues, we developed THINGS, a database of 1854 representatively-sampled object concepts, more than 26,000 high quality object images, and diverse metadata.
The database serves as the basis for the THINGS initative, a global initiative to provide a common framework for studying object representations in humans, animal models, and artificial intelligence research. Read the paper accompanying the database here. To download THINGS, visit the project page here.