Natural Language Dialogue group

The USC SmartBody Annotation Retrieval Dataset(SARD) is a collection of 274 animation videos of the SmartBody character, Brad. Every video is annotated with anywhere between 1 to 6 1-word long concepts, such as "Arms", "Legs", etc. The total number of concepts is 51.

The dataset consists of the videos, the annotation labels, the extracted features, and 3d coordinates of the joints.

The primary intention of making the dataset public is to facilitate multimedia annotation/retrieval research in the community. However, people are welcome to use it for other research related tasks as well! Please make sure that you do follow the recommended licensing and citation instructions.

For details about how to use the dataset, licensing, citation, please read the README file in the folder.

For a detailed report on the various approaches tried, please check the following document:
http://nld.ict.usc.edu/group/sites/default/files/AnnotRet_Doc_Compressed...