SalSOD dataset

This repository provides the SalSOD dataset. SalSOD is a large-scale salient object ranking dataset of 24,373 images, with multiple annotations on the salient objects inside images: 1) semantic segmentation masks and bounding boxes, 2) saliency values and saliency rankings, and 3) scene graphs of objects.

The dataset establishment process are illustrated in the following figure.

Example - saliency annotation

Example - scene graph annotation

Download dataset

SalSOD database can be downloaded from DropBox(key:salrank) or BiaduPan(key:hejz). Please feel free to contact us by clicking here (email: minglangqiao@buaa.edu.cn) if you have any problems.

TODO list

• release the dataset [part1]
• release the dataset [part2]
• release the usuage code