Autoset-1K

A dataset constructed from large scale Click-through Logs automaticly by using Deep Neural Network.

Overview

t-SNE visualization of learned word representations: (high resolution)

t-SNE visualization of learned image representations: (high resolution)

Download

1. Autoset-1K

  • For each category, a set of image ID in Clickture-Full is given. You can extract Autoset-1K image dataset according to our given image ID from Click-Full which contains 40 million of images.
  • You must accept the enclosed License Terms of Bing-MSR Image Annotation Challenge Data in order to use Click-Full dataset. You are not allowed to further distribute Autoset-1K and Click-Full dataset.

    • 2. Visual based Word Embeddings

    3. KNN results of Visual based Word Embeddings