CAT is a large-scale dataset built for co-saliency detection. It's an extension of the existing salient object detection (SOD) datasets with considerations of grouping and object co-occurrence.

  • Grouping: .
  • Object co-occurrence: .

Basic Statistics

CAT has a total of 74,250 images, evenly distributed in 9 sub-sets. These sub-sets are divided by the following two indicators:

  • SIZE ('Tiny', 'Middle', 'Large'): Relative size proportional to the size of the input image. Here we choose 'Tiny'=0.2, 'Middle'=0.3, and 'Large'=0.4.
  • NUM ('Double', 'Triple', 'Quadruple'): The number of sampled object(s) put in the image. Here we choose 'Double'=1, 'Triple'=2, 'Quadruple'=3.


You are free to choose and download whatever combincation you need.

'Tiny' 'Middle' 'Large'
'Double' [Google Drive] [Google Drive] [Google Drive]
'Triple' [Google Drive] [Google Drive] [Google Drive]
'Quadruple' [Google Drive] [Google Drive] [Google Drive]


If you find this work useful, please cite the following papers:

title={Learning to Detect Salient Objects with Image-level Supervision},
author={Wang, Lijun and Lu, Huchuan and Wang, Yifan and Feng, Mengyang and Wang, Dong, and Yin, Baocai and Ruan, Xiang}, 


Nunc lacinia ante nunc ac lobortis. Interdum adipiscing gravida odio porttitor sem non mi integer non faucibus ornare mi ut ante amet placerat aliquet. Volutpat eu sed ante lacinia sapien lorem accumsan varius montes viverra nibh in adipiscing blandit tempus accumsan.


Lingdong Kong

Nanyang Technological University

Junhao Liu

SIAT, Chinese Academy of Sciences

Yao Chen

Advanced Digital Sciences Center

Prakhar Ganesh

Advanced Digital Sciences Center

Tan Wang

Nanyang Technological University

Hanwang Zhang

Nanyang Technological University