BR-Gen
收藏资源简介:
BR-Gen是一个大规模、高质量的图像数据集,包含150,000张经过局部篡改的图像,涵盖了多样化的场景感知注释。该数据集由教育部多媒体可信感知与高效计算重点实验室(厦门大学)构建,通过全自动的感知-创建-评估管道,确保了图像的语义完整性和视觉真实性。数据集针对以往数据集中忽视的‘事物’和‘背景’类别,如天空、地面、墙壁、草地和植被等,拓宽了局部篡改图像的范畴。
BR-Gen is a large-scale, high-quality image dataset containing 150,000 locally tampered images with diverse scene-aware annotations. It was constructed by the Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education, Xiamen University. A fully automatic perception-creation-evaluation pipeline is adopted to ensure the semantic integrity and visual authenticity of the images. Moreover, this dataset expands the scope of locally tampered images by covering the 'object' and 'background' categories that were neglected in previous datasets, such as sky, ground, wall, grassland and vegetation.

- 1Zooming In on Fakes: A Novel Dataset for Localized AI-Generated Image Detection with Forgery Amplification Approach中国教育部多媒体可信感知与高效计算重点实验室,厦门大学 · 2025年



