New BG-REAL benchmark targets background image manipulation detection

Researchers have introduced BG-REAL, a new benchmark designed for detecting and localizing background manipulations in images. This benchmark is built using real-world data from Open Images V7 and includes 7,000 processed samples across six edit families, along with matched authentic controls. Initial evaluations of three existing baselines—TruFor, MVSS-Net, and HiFi-Net—revealed significant false-positive rates, indicating a shared risk of misclassifying re-encoded authentic images as manipulated. AI

IMPACT Introduces a specialized benchmark for image forensics, potentially improving the robustness of manipulation detection models against background edits.

RANK_REASON Publication of a new academic benchmark dataset and paper. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• BG-REAL
• Buğra Alperen Uluırmak
• HiFi-Net
• MVSS-Net
• Open Images V7
• Trufor

• paper
• other

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