Datasets:

nobg
/

DUT-OMRON

license: cc-by-nc-4.0
pretty_name: DUT-OMRON
task_categories:
  - image-segmentation
tags:
  - saliency
  - salient-object-detection
size_categories:
  - 1K<n<10K

DUT-OMRON

The DUT-OMRON dataset is a widely used benchmark for salient object detection / saliency prediction. It contains 5,168 high-quality natural images, each paired with pixelwise ground-truth saliency masks.

image: the source photograph (JPEG, RGB).
mask: the pixelwise binary ground-truth saliency mask (PNG, grayscale L).

Images and masks are matched by filename stem. This release provides a single test split (the dataset is a single benchmark set).

Source & credit

Data from the DUT-OMRON dataset, distributed at https://saliencydetection.net/dut-omron/. All credit to the original authors. If you use this dataset, please cite the original DUT-OMRON paper:

C. Yang, L. Zhang, H. Lu, X. Ruan, M.-H. Yang. "Saliency Detection via Graph-Based Manifold Ranking." CVPR 2013.

License

Provided for academic / research use. No explicit SPDX license is stated by the source; distributed here under cc-by-nc-4.0 (non-commercial).