krutrim-ai-labs/IndicPope
收藏Hugging Face2025-03-05 更新2025-04-08 收录
下载链接：
https://hf-mirror.com/datasets/krutrim-ai-labs/IndicPope
下载链接
链接失效反馈官方服务：
资源简介：
IndicPope是一个用于评估大型视觉语言模型在视觉问答任务上的数据集，特别针对简单的是非问题（例如，图片中是否有车？）。该数据集基于POPE，并使用负采样技术来测试随机、流行和对抗性设置下的视觉语言模型中的对象幻觉。数据集包含8.91k个样本，涵盖12种印度语言和英语。
IndicPope is a dataset designed for evaluating Large Vision-Language Models (LVLMs) on Visual Question Answering (VQA) tasks, particularly for simple Yes-or-No questions about objects in images (e.g., Is there a car in the image?). Built upon POPE, it employs negative sampling techniques to test hallucination in vision-language models under Random, Popular, and Adversarial settings. The dataset consists of 8.91k samples spanning 12 Indic languages and English.
提供机构：
krutrim-ai-labs
