memmelma/peek_vqa
收藏Hugging Face2025-10-03 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/memmelma/peek_vqa
下载链接
链接失效反馈官方服务：
资源简介：
PEEK VQA数据集包含200万个图像问答对，用于微调机器人操作中的视觉语言模型PEEK VLM。该数据集由多个机器人操作数据集构成，如bridge_v2、droid等，经过特定的标注流程生成。数据集以.json格式存储，包含会话信息和图像引用，用于指导机器人执行特定任务时的动作路径和关注区域。
The PEEK VQA dataset consists of 2M image-QA pairs for fine-tuning the PEEK VLM model in robotic operations. It is composed of multiple robotics datasets like bridge_v2, droid, etc., annotated through a specific pipeline. The dataset is stored in .json format, containing conversation information and image references, used to guide the robot in action trajectories and focus areas for performing specific tasks.
提供机构：
memmelma
