Paper finds human damage assessment accuracy varies by imagery source

A new paper investigates how human annotators and reviewers perform when assessing aerial damage from different sources like drones, crewed aircraft, and satellites. The study found that initial annotations required significant revisions by a final committee, with lower-resolution sources like satellite imagery needing the most correction. Even after a single reviewer pass, substantial disagreements persisted, particularly with satellite data, suggesting that uniform review allocation may not be optimal for multi-source datasets. AI

IMPACT Highlights challenges in data annotation for AI models, particularly concerning the reliability of different imagery sources for training.

RANK_REASON Academic paper published on arXiv [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.AI →

• arXiv
• crewed aviation
• Hugging Face
• Satellite
• unmanned aerial vehicle

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →