Ivy-Fake: A New Benchmark for Explainable AIGC Detection
Researchers have introduced Ivy-Fake, the first large-scale multimodal benchmark for explainable detection of AI-generated content (AIGC) in images and videos. The benchmark addresses two major limitations in current detection methods: the lack of multidimensional explainable datasets and insufficiently fine-grained interpretability in prior MLLM-based forgery detectors. Ivy-Fake comprises over 106,000 richly annotated training samples and 5,000 manually verified evaluation samples. It aims to enhance the explainability and trustworthiness of detectors by providing detailed annotations beyond binary labels, enabling better localization and reasoning about forgeries. The work is published on arXiv under paper number 2506.00979.
Key facts
- Ivy-Fake is the first large-scale multimodal benchmark for explainable AIGC detection.
- It includes over 106,000 training samples and 5,000 evaluation samples.
- The benchmark addresses limitations in existing datasets like WildFake and GenVideo.
- It improves upon prior MLLM-based detectors like FakeVLM by offering more fine-grained interpretability.
- The research is published on arXiv with paper number 2506.00979.
Entities
Institutions
- arXiv