OmniGen2/OmniContext

OmniContext是一个用于评估模型上下文生成能力的基准数据集，包含多种任务类型，如图像-文本到图像、图像到图像、任意到任意等。数据集由多样化的输入图像和相应的用户指令组成，支持使用GPT-4.1进行度量的评估。每个样本都包括任务类型、指令、输入图像列表、指令语言和唯一标识符。

OmniContext is a benchmark dataset for evaluating models in-context generation capabilities, incorporating various task types such as image-text-to-image, image-to-image, any-to-any, etc. The dataset consists of diverse input images paired with corresponding user instructions and supports GPT-4.1-based metric evaluation. Each sample includes the task type, instruction, a list of input images, instruction language, and a unique identifier.