camobench/CamoBench
收藏资源简介:
CamoBench数据集是一个用于评估AI生成图像中伪装对象检测和编辑性能的数据集。它包含背景图像和伪装对象图像,其中对象图像有多个变体,如原始图像、黑色背景图像、深度图像等。数据集还提供了背景和对象的类别、描述以及生成提示和指令,支持条件生成、编辑和文本到图像任务。仓库结构包括生成的图像、原始图像、提示文件和评分文件,适用于多模型比较和性能评估。
CamoBench is a dataset designed to evaluate the performance of camouflaged object detection and editing in AI-generated images. It comprises background images and camouflaged object images, where the object images come in multiple variants such as original images, black-background images, depth images, and more. The dataset also provides categories and descriptions for both backgrounds and objects, alongside generation prompts and instructions, enabling support for conditional generation, editing, and text-to-image tasks. The dataset's repository structure includes generated images, raw images, prompt files, and score files, making it suitable for multi-model comparison and performance evaluation.





