Graph200K
收藏资源简介:
Graph200K是一个图结构的多任务数据集,由南开大学计算机科学与技术学院等机构构建。该数据集对Subject200K数据集进行增强,为每张图像添加了49种不同任务的标注,覆盖了条件生成、图像修复、图像编辑、IP保持和风格转换等五个元任务。通过任务标注的组合,数据集支持构建多样化的相关任务,促进了模型在任务间的知识共享和迁移性学习,增强了模型的泛化能力。
Graph200K is a graph-structured multi-task dataset constructed by the College of Computer Science and Technology of Nankai University and other institutions. This dataset is an enhanced version of the Subject200K dataset, which adds annotations for 49 distinct tasks to each image, covering five meta-tasks including conditional generation, image inpainting, image editing, identity preservation, and style transfer. By combining these task annotations, the dataset supports the construction of diverse related tasks, facilitating knowledge sharing and transfer learning across tasks for models and enhancing the generalization capability of the models.

- 1VisualCloze: A Universal Image Generation Framework via Visual In-Context Learning南开大学计算机科学与技术学院, 北京邮电大学, 清华大学, 上海人工智能实验室, 香港中文大学 · 2025年



