GIM
收藏资源简介:
GIM数据集是由华为诺亚方舟实验室创建的大规模生成式图像操作检测与定位基准,包含超过一百万对AI操作图像和真实图像。该数据集内容丰富,涵盖广泛的图像类别,并采用多种先进的生成模型进行图像操作。创建过程中,利用了强大的扩散模型和SAM技术,确保了数据集的高保真度和多样性。GIM数据集主要应用于AI生成内容的安全性检测,旨在解决图像操作检测与定位的挑战,推动相关技术的发展。
The GIM dataset is a large-scale benchmark for generative image manipulation detection and localization developed by Huawei Noah's Ark Lab. It contains over one million pairs of AI-manipulated images and authentic real-world images, with rich content covering a wide range of image categories. The image manipulations in the dataset are generated using multiple state-of-the-art generative models. During its creation, robust diffusion models and the Segment Anything Model (SAM) were leveraged to ensure the dataset's high fidelity and diversity. Primarily applied for safety detection of AI-generated content, the GIM dataset aims to address the challenges in image manipulation detection and localization, and promote the advancement of related technologies.

- 1GIM: A Million-scale Benchmark for Generative Image Manipulation Detection and Localization华为诺亚方舟实验室 · 2024年



