---
license: cc-by-nc-4.0
pretty_name: DUTS
task_categories:
- image-segmentation
tags:
- saliency-detection
- salient-object-detection
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---

# DUTS

DUTS is a large-scale saliency detection (salient object detection) dataset. It
contains a training set, **DUTS-TR** (10,553 images), and a test set,
**DUTS-TE** (5,019 images). Each image is paired with a binary ground-truth
saliency mask.

## Splits

| Split | Source   | Rows   |
|-------|----------|--------|
| train | DUTS-TR  | 10,553 |
| test  | DUTS-TE  | 5,019  |

## Columns

- `image`: the RGB input image (`datasets.Image`).
- `mask`: the ground-truth saliency mask (`datasets.Image`, single channel).

Image and mask are matched by filename stem.

## License

Released for academic / research use. No explicit SPDX license is provided by
the authors; this mirror is published under `cc-by-nc-4.0`. See
<https://saliencydetection.net/duts/> for the original terms.

## Credits

Source: <https://saliencydetection.net/duts/>

Paper: Lijun Wang, Huchuan Lu, Yifan Wang, Mengyang Feng, Dong Wang, Baocai Yin,
Xiang Ruan. *Learning to Detect Salient Objects with Image-level Supervision.*
CVPR 2017.