MSTF
收藏资源简介:
MSTF是由中山大学构建的全球首个大规模多场景对话人脸数据集,旨在填补对话人脸生成检测领域的数据集空白。该数据集包含超过14万条音频和视频数据,涵盖22种伪造技术、11种生成场景和20多种语义场景,更贴近实际应用场景。数据集的创建过程结合了多种图像、音频和视频数据源,通过模拟多种生成场景,确保数据集的多层次一致性。MSTF主要应用于对话人脸生成检测领域,旨在解决现有深度伪造检测方法在对话人脸视频中的局限性,推动高精度检测技术的发展。
MSTF is the world's first large-scale multi-scenario conversational face dataset constructed by Sun Yat-sen University, aiming to fill the gap of datasets in the field of conversational face generation detection. This dataset contains over 140,000 audio and video data samples, covering 22 types of forgery technologies, 11 generation scenarios and more than 20 semantic scenarios, making it more aligned with real-world application scenarios. The construction of the dataset integrates multiple image, audio and video data sources, and simulates diverse generation scenarios to ensure the multi-level consistency of the dataset. MSTF is primarily applied in the field of conversational face generation detection, aiming to address the limitations of existing deepfake detection methods in conversational face videos and promote the development of high-precision detection technologies.

- 1GLCF: A Global-Local Multimodal Coherence Analysis Framework for Talking Face Generation Detection中山大学 · 2024年



