CC12M (Conceptual 12M)

概念12 M (CC12M)，具有约1200万个图像文本对的数据集，旨在用于视觉和语言预训练。与概念字幕 (CC3M) 相比，它更大，涵盖了更多的视觉概念集，该概念集广泛用于图像字幕模型的预训练和端到端训练。

Conceptual 12M (CC12M) is a dataset consisting of approximately 12 million image-text pairs, designed for vision-and-language pre-training. Compared to Conceptual Captions (CC3M), it has a larger scale and covers a more extensive set of visual concepts, which has been widely used for pre-training and end-to-end training of image captioning models.

搜集汇总

背景与挑战

背景概述

CC12M是一个由Google Research于2021年发布的大规模图像文本对数据集，包含约1200万个样本，旨在用于视觉和语言预训练。它比前代CC3M更大，覆盖了更广泛的视觉概念集，特别关注长尾视觉概念的识别，适用于图像字幕模型的预训练和端到端训练。