RVMS-Bench: 10,000 Bilingual Video Search and Localization Queries

RVMS-Bench is a benchmark for real-world video search and moment localization developed by Tencent in 2026. It contains between 1,000 and 10,000 text-based query annotations and metadata designed for agent-based retrieval frameworks. This specific repository provides the search paradigm metadata but excludes raw video assets and ground-truth keyframes.

Use Cases

• Training search agents using the text-based query annotations
• Implementing moment localization logic using the search paradigm metadata
• Evaluating bilingual retrieval systems using the English and Chinese language labels

Strengths

• Bilingual support with English and Chinese query annotations
• 1,000 to 10,000 annotated records
• Apache 2.0 licensed for open research use

Limitations

• Missing ground-truth keyframe images and evaluation assets
• Relatively small scale of under 10,000 records
• Requires external data sourcing for complete model evaluation

Provenance

Source

Tencent (Arxiv 2602.10159)

Collection Method

Expert-annotated text queries and metadata for video search

Freshness

Last updated March 2026.