LVIS_Instruct
收藏资源简介:
该数据集是一个多模态数据集,包含图像及对应的双语对话文本。每个样本由三个主要部分构成:1) 图像数据(images字段);2) 原始文本对话(original_text字段),以列表形式组织,包含用户(user)和助手(assistant)的对话轮次;3) 翻译后的文本对话(translated_text字段),结构与原始文本相同,同样包含用户和助手的对话内容。数据集规模为100个训练样本,总数据量约为49.48MB。从数据结构推断,该数据集适用于多模态任务,如图像描述生成、视觉问答、对话系统构建,以及跨语言的自然语言处理任务,如机器翻译或双语对话生成。数据以用户-助手对话对的形式呈现,暗示其可能用于指令遵循或对话代理的训练与评估。
This is a multimodal dataset containing images and their corresponding bilingual dialogue transcripts. Each sample comprises three core components: 1) Image data, stored in the `images` field; 2) Original textual dialogue, stored in the `original_text` field and structured as a list containing dialogue turns between the user and the assistant; 3) Translated textual dialogue, stored in the `translated_text` field, which adopts the identical structure as the original dialogue and also covers conversation content between the user and the assistant. The dataset consists of 100 training samples, with an overall data size of approximately 49.48 MB. Given its data structure, this dataset is applicable to multimodal tasks such as image captioning, visual question answering (VQA), dialogue system development, as well as cross-lingual natural language processing tasks including machine translation and bilingual dialogue generation. The data is formatted as user-assistant dialogue pairs, indicating its potential use cases for training and evaluating instruction-following models or dialogue agents.




