WebVision 1.0+2.0 Image Classification Dataset

The WebVision dataset is a large-scale web image dataset containing more than 2.4 million images crawled from Flickr and Google Image Search. The corresponding text information of the images (such as title, user tags or description) is also provided as additional meta information. The dataset provides a validation set of 50,000 images (50 per category). The dataset can be used to promote research on learning visual representations from noisy web data.

1,000 images with the same concept as the ILSVRC 2012 dataset are used for search, making existing methods directly applicable to research and comparable to models trained on the ILSVRC 2012 dataset, while also making it possible to study dataset bias issues in large-scale scenarios.

This dataset includes versions 1.0 and 2.0.