Friends-MMC
收藏资源简介:
Friends-MMC是一个多模态多方对话数据集,由北京大学王选计算机研究所创建,基于著名电视剧《老友记》构建。该数据集包含超过24,000条独特的话语,每条话语都与视频上下文配对,并标注了说话者、面部名称和边界框。数据集的创建过程包括视频预处理、面部检测与标注、对话会话选择等步骤。Friends-MMC旨在促进多模态多方对话理解的研究,特别是在对话发言人识别和对话响应预测任务中,解决现实场景中的复杂对话理解问题。
Friends-MMC is a multimodal multi-party dialogue dataset developed by the Wangxuan Institute of Computer Technology at Peking University, based on the iconic TV series *Friends*. This dataset contains more than 24,000 unique utterances, each paired with corresponding video context, and is annotated with speaker identities, facial names and bounding boxes. The construction process of Friends-MMC includes several steps such as video preprocessing, face detection and annotation, and dialogue session selection. This dataset aims to promote research on multimodal multi-party dialogue understanding, particularly in the tasks of dialogue speaker recognition and dialogue response prediction, to address complex dialogue understanding challenges in real-world scenarios.

- 1Friends-MMC: A Dataset for Multi-modal Multi-party Conversation Understanding北京大学王选计算机研究所 · 2024年



