Benchmark RadarT-IMPACT
T-IMPACT evaluates models on severity-aware detection of manipulated news-style image-text pairs, with 98,786 examples covering pristine, image-only, text-only, and joint manipulations, alongside calibrated continuous severity scores, coarse labels, and grounding metadata.
- Released
- 2026-06-21
- Readiness
- Paper only
- Primary field
- Robotics & Autonomous Systems
Why it matters
Existing multimodal manipulation benchmarks focus on authenticity or manipulation type, lacking graded impact severity. T-IMPACT fills this gap with a calibrated continuous severity signal, enabling evaluation of models' ability to judge contextual impact, which is essential for mitigating persuasive misinformation.
Motivation
Recent advances in vision-language models and generative editing systems have made it increasingly easy to produce persuasive multimodal misinformation by altering images, text, or both jointly.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.