GRADE
收藏资源简介:
GRADE是由上海交通大学等机构联合构建的首个面向学科知识推理的图像编辑基准数据集,涵盖数学、物理、化学等10个学科领域的520个精细标注样本。数据集包含输入图像、文本编辑指令和真实编辑结果的三元组结构,数据来源于开放教材和专业参考资料,并经过多轮专家验证。该数据集旨在评估多模态模型在学科知识引导下的复杂图像编辑能力,为科研辅助、教育工具等场景提供基准测试平台。
GRADE is the first benchmark dataset for image editing guided by disciplinary knowledge reasoning, jointly constructed by Shanghai Jiao Tong University and other institutions. It contains 520 finely annotated samples spanning 10 disciplines including mathematics, physics, chemistry and other fields. The dataset adopts a triplet structure consisting of input images, text editing instructions and ground-truth edited results. The data is sourced from open textbooks and professional reference materials, and has undergone multiple rounds of expert validation. This dataset aims to evaluate the complex image editing capabilities of multimodal models guided by disciplinary knowledge, providing a benchmark platform for scenarios such as scientific research assistance and educational tools.

- 1GRADE: Benchmarking Discipline-Informed Reasoning in Image Editing上海交通大学; 华南理工大学; 复旦大学; 香港中文大学; 中国科学技术大学 · 2026年



