Benchmark RadarAI BENCHMARK PROFILE
BeyondMasks
Evaluates causally consistent video object removal using paired synthetic and real videos with aligned references, and includes the CORE metric for after-effect consistency.
- Released
- 2026-08-20
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Introduces a benchmark that checks whether removed objects' induced physical effects are also eliminated, addressing a key gap in video editing evaluation beyond masked region fidelity.
Motivation
Recent advances in generative video models have significantly improved visual realism in video object removal, yet evaluation protocols still focus on masked region fidelity, treating removal as local inpainting.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.