MMDialog is a large-scale multimodal open-domain dialogue dataset that contains 1.08 million complete dialogue sessions, more than 4,000 dialogue topics, and 1.53 million non-repeated images. Each dialogue session has an average of 2.59 images and can be located at any position in the dialogue process.

MMDialog's rich and authentic human conversation content is collected from an English online social platform (Note: this process fully complies with the platform's regulations on the collection and sharing of academic research data, and user privacy has been anonymized and data information has been encrypted).