Researchers have developed a novel training-free method for attributing AI-generated videos to their specific sources. This approach treats video attribution as an instance retrieval task, employing a pipeline that includes adapted orthogonal color transformation, multi-scale quantized residual generation, and temporal-semantic aggregation to capture generative artifacts across video frames. Tested on the GenVidBench benchmark, the method demonstrated strong performance in both detection and attribution, achieving a Rank-1 accuracy of 20.5% and a mean Average Precision of 16.6%, outperforming existing state-of-the-art techniques. AI
IMPACT Enhances forensic capabilities for identifying the origins of AI-generated video content, crucial for combating misuse.
RANK_REASON Academic paper detailing a new method for AI-generated video attribution. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
