Benchmark RadarSciFigPlag-Bench
SciFigPlag-Bench evaluates provenance-aware reasoning for scientific figure plagiarism detection. It includes 2,582 positive and 2,541 negative pairs, with tasks for pairwise detection, source attribution, reuse-type classification, and reuse correspondence localization.
- Released
- 2026-07-31
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Figure plagiarism is underexplored and general similarity benchmarks do not assess provenance. SciFigPlag-Bench provides a factorized taxonomy and diagnostic tasks to evaluate multimodal models on fine-grained provenance reasoning, aiding research integrity.
Motivation
Scientific figures often encode the visual evidence behind scientific findings, yet figure plagiarism remains underexplored as a benchmarked multimodal evaluation problem.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.