Visual Genome Dataset V1.2 Large-scale Image Semantic Understanding Dataset

* This dataset supports online use.Click here to jump.

Visual Genome is a dataset related to semantic information of images and image content. Compared with the ImageNet image annotation dataset, it has richer semantic information. It is used to expand artificial intelligence applications based on images and semantic information.

The dataset currently contains 108,249 images, 4.2 million region content descriptions, 1.7 million image content questions and answers, 2.1 million object cases, 1.8 million attributes, and 1.8 million relationships.

The dataset was first released by Stanford University in 2015, version 1.2 was released in 2016, and version 1.4 was released in 2017.