TVQA+: Spatio-Temporal Grounding for Video Question Answering

TVQA+ provides spatio-temporal grounding labels for video question answering tasks. Developed by researchers for ACL 2020, the dataset facilitates multi-modal reasoning by linking natural language questions to specific video frames and regions.

Use Cases

• Training models for spatio-temporal grounding using bounding box annotations
• Developing multi-modal QA systems that align subtitle text with visual frames
• Benchmarking visual reasoning by localizing objects mentioned in question-answer pairs

Strengths

• Provides spatio-temporal grounding via bounding box annotations
• Integrates video frames with subtitle text and QA pairs
• Peer-reviewed methodology from ACL 2020

Limitations

• Restricted to television show environments which may not generalize to other domains
• Static dataset with no updates since 2022

Provenance

Source

ACL 2020 Publication: TVQA+: Spatio-Temporal Grounding for Video Question Answering

Collection Method

Annotated from television show frames and subtitles

Freshness

Last updated in October 2022; static research dataset.