CoCoVideo: The High-Quality Commercial-Model-Based Contrastive Benchmark for AI-Generated Video Detection
Researchers have introduced CoCoVideo-26K, a new benchmark dataset designed to improve the detection of AI-generated videos, particularly those created by high-fidelity commercial models. The dataset features semantically aligned real and fake video pairs from 13 mainstream commercial generators, addressing limitations of existing datasets that often use lower-quality open-source models or suffer from visible watermarks. Alongside the dataset, the team developed CoCoDetect, a framework that combines contrastive learning with multimodal large language model (MLLM) inference to achieve state-of-the-art performance in detecting realistic AI-generated videos. AI
IMPACT Improves AI-generated video detection capabilities, crucial for combating misinformation and ensuring authenticity.