MM-Hallu/NOPE

NOPE是基于COCO val2017的负样本探测评估数据集，包含14,856个是/否问题，覆盖三种负采样策略（随机、流行、对抗），每种策略有4,952个问题。问题格式为图像中是否有{object}?，用于评估模型对图像中对象存在的判断能力。数据集包含图像、图像ID、问题ID、问题文本、探测对象、答案（是/否）、标签（1表示存在/0表示不存在）、采样策略类型和图像中实际存在的对象列表等字段。

Negative Object Probing Evaluation on COCO val2017. 14,856 yes/no questions across three negative sampling strategies (random, popular, adversarial), each with 4,952 questions. Questions are in the format Is there a {object} in the image? to evaluate models ability to judge object presence. Dataset includes fields: image, image_id, question_id, question, object, answer (yes/no), label (1 for present/0 for absent), split_type (sampling strategy), and present_objects (actual objects in image).