PARROT-360V
收藏资源简介:
PARROT-360V数据集由红块人工智能创建,旨在评估视觉语言模型在复杂视觉推理任务中的表现。该数据集包含2487个精心挑选的视觉谜题,每个谜题包含多个特征,如日期、PDF文件路径、问题截图、答案截图等。数据集的创建过程涉及从互联网上抓取Jumble谜题,并提取额外的特征以获取地面真值标签。PARROT-360V数据集的应用领域主要集中在视觉语言模型的评估和改进,旨在解决当前模型在复杂、多步骤推理任务中的局限性。
The PARROT-360V dataset, created by Redblock AI, is designed to evaluate the performance of vision-language models on complex visual reasoning tasks. This dataset comprises 2487 carefully selected visual puzzles, each containing multiple features such as dates, PDF file paths, screenshots of questions, and screenshots of answers, among others. The dataset creation process involved scraping Jumble puzzles from the internet and extracting additional features to obtain ground-truth labels. The application scenarios of the PARROT-360V dataset mainly focus on the evaluation and improvement of vision-language models, with the goal of addressing the limitations of current models in complex, multi-step reasoning tasks.




