← Yalong Bai

Autoset-1K

An image dataset constructed automatically from large-scale click-through logs using deep neural networks.

Yalong Bai, Kuiyuan Yang, Wei Yu, Chang Xu, Wei-Ying Ma, Tiejun Zhao. Automatic Image Dataset Construction from Click-through Logs Using Deep Neural Network. ACM Multimedia, 2015 pdf

Overview

Overview of the Autoset-1K construction pipeline
t-SNE visualization of learned word representations
t-SNE visualization of learned word representations. High-resolution version (6000×3600).
t-SNE visualization of learned image representations
t-SNE visualization of learned image representations. High-resolution version (6000×3600, 6.5 MB).

Download

  1. Autoset-1K

    For each category, a set of image IDs from Clickture-Full is provided. You can extract the Autoset-1K image dataset from Clickture-Full, which contains 40 million images, using the given image IDs.

    You must accept the License Terms of the MSR-Bing Image Retrieval Challenge data in order to use the Clickture-Full dataset. You are not allowed to further distribute Autoset-1K or Clickture-Full.

  2. Visual-based Word Embeddings

  3. KNN Results of Visual-based Word Embeddings