linjieli222/spatial-imaginative-token-pt-eval-ai2thor
收藏资源简介:
该数据集是一个多模态视觉问答数据集,包含三个配置:td_ego_dir、td_path和td_path_arrow。每个配置由训练集组成,具有文本和图像特征。文本特征包括问题ID、问题、答案和选项列表;图像特征包括俯视图图像、自我中心图像列表、侧视图图像和侧视图描述。数据集可能用于导航或空间推理任务,涉及从不同视角(如俯视图和第一人称视图)理解环境并回答问题。具体来说,td_ego_dir可能关注自我中心方向判断,td_path可能涉及路径识别,td_path_arrow可能添加了箭头指示以增强路径理解。数据规模较小,每个配置的样本数在113到171之间。
This dataset is a multimodal visual question answering dataset with three configurations: td_ego_dir, td_path, and td_path_arrow. Each configuration consists of a training split with text and image features. Text features include question ID, question, answer, and a list of choices; image features include a top-down image, a list of ego-centric images, a side-view image, and a side-view description. The dataset is likely designed for navigation or spatial reasoning tasks, involving understanding environments from different perspectives (e.g., top-down and first-person views) to answer questions. Specifically, td_ego_dir may focus on ego-centric direction judgment, td_path may involve path identification, and td_path_arrow may incorporate arrow indicators for enhanced path understanding. The dataset is relatively small, with sample sizes ranging from 113 to 171 per configuration.




