Benchmark RadarCapRiCorn-1K
CapRiCorn-1K evaluates video captioning quality and subject referential consistency across long videos (15s-10min) with audiovisual and visual-only settings. It uses LLM judge to compute accuracy, coverage, and referential consistency metrics based on manual annotations.
- Released
- 2026-06-20
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Existing benchmarks focus on short videos and overall caption quality, lacking evaluation of subject referential consistency over long horizons. CapRiCorn-1K's metrics correlate with downstream understanding and generation performance, offering practical value for selecting captioning models.
Motivation
Accurate and comprehensive video captions with consistent subject references are critical for downstream understanding and generation tasks.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.