Forensics Adapter enhances CLIP for generalizable face forgery detection

Researchers have developed a new method called Forensics Adapter, which modifies the CLIP model to become a more effective and generalizable face forgery detector. Unlike previous approaches that treated CLIP solely as a feature extractor, this new adapter learns specific traces of forged faces, such as blending boundaries. The adapter works alongside CLIP, preserving its versatility and ensuring strong performance across multiple datasets with a relatively small number of trainable parameters. An extended version, Forensics Adapter++, further improves performance by incorporating textual modality through a novel prompt learning strategy. AI

IMPACT This research introduces a more efficient and generalizable method for detecting AI-generated or manipulated images, which could have implications for content authenticity and security.

RANK_REASON The cluster contains an academic paper detailing a new method for face forgery detection. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Forensics Adapter
• Yuezun Li

• paper
• model release

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