ReVision Dataset
收藏资源简介:
ReVision数据集是一个专门为视觉指令重写任务设计的多模态数据集,由德克萨斯大学奥斯汀分校信息学院和耶鲁大学统计与数据科学系创建。该数据集包含超过39000个示例,跨越14个领域,由图像、原始指令和重写指令组成的三元组构成。数据集旨在将复杂的多模态交互转化为纯文本指令,以便在保持隐私的同时,在设备上执行。数据来源于公共可用的学术数据集,通过GPT-4模型生成和验证指令,并经过人工评估以确保可靠性。
ReVision Dataset is a multimodal dataset specifically tailored for the visual instruction rewriting task, developed by the School of Information at The University of Texas at Austin and the Department of Statistics and Data Science at Yale University. It contains over 39,000 examples spanning 14 domains, structured as triplets consisting of images, original instructions, and rewritten instructions. The dataset aims to convert complex multimodal interactions into plain-text instructions, enabling on-device execution while preserving privacy. The data is sourced from publicly available academic datasets, with instructions generated and verified via the GPT-4 model, and subjected to human evaluation to ensure reliability.

- 1ReVision: A Dataset and Baseline VLM for Privacy-Preserving Task-Oriented Visual Instruction Rewriting德克萨斯大学奥斯汀分校信息学院, 耶鲁大学统计与数据科学系 · 2025年



