SUN Attribute Database:
Discovering, Annotating, and Recognizing Scene Attributes

Hays Lab | 2011



In this paper we present the first large-scale scene attribute database. First, we perform crowd-sourced human studies to find a taxonomy of 102 discriminative attributes. Next, we build the "SUN attribute database'' on top of the fine-grained SUN categorical database. Our attribute database spans more than 700 categories and 14,000 images and has potential for use in high-level scene understanding and fine-grained scene recognition. We use our dataset to train attribute classifiers, and evaluate how well these relatively simple classifiers can recognize a variety of attributes related to materials, surface properties, lighting, functions and affordances, and spatial envelope properties.


Genevieve Patterson, Chen Xu, Hang Su, James Hays. The SUN Attribute Database: Beyond Categories for Deeper Scene Understanding. IJCV 2014.
paper, Bibtex
Genevieve Patterson, James Hays. SUN Attribute Database: Discovering, Annotating, and Recognizing Scene Attributes. Proceedings of CVPR 2012.
paper, Bibtex

SUN Attribute Dataset

This dataset includes the 102 attribute labels x 3 worker annotations for each of the 14340 images included. The subset of images from the SUN Dataset used in this project are also available for download from the link below. Users can also download the SUN dataset images used in this project at the SUN Database website.

Attribute Labels(532 KB) including list of images used from SUN dataset.
AttributeDB Images(1.7 GB) including all 14340 images used in the SUN Attribute dataset.
AttributeDB Images' Features(240.3 MB) including all of the pre-calculated image features used in the SUN Attribute DB paper.
AttributeDB Test/Train splits(34.2 MB) including all the test/train splits used to train and test the SVMs whose performance is reported in the paper.

New Code Release v2.1! (Updated 2/2/2013)

This new release is a large speed up. Attribute features are calculated per image in ~12sec on a 3.4GHZ AMD Phenom(tm) II X4 965 Processor with cache size 512 KB and 32GB of memory. Average memory usage is 4GB per Matlab instance.

To classify scene attributes in novel images, download our attribute classifiers. The packages below include pre-computed classifiers and the code for generating and using them. The thumbnail to the left shows the AP Scores for each attribute classifier, click for enlarged version.

GitHub Repo

AttributeDB pre-calculated SVM models and kernel matricies (1.1 GB)(tar.gz version)

AttributeDB pre-calculated SVM models and kernel matricies (1.1 GB)(zip version)

AttributeDB classifier code for Matlab (9.5 KB)