I would like to understand reasons for different image datasets used for salient object detection: their difficulties, their origins and likely future datasets. Does there exist any work explaining them such their progression and characteristics over time?
$\begingroup$
$\endgroup$
Datasets explicitly mentioning saliency
- MSRA
- SOD based on BSD Berkeley segmentation dataset
Apparently more general datasets (not researched in detail)
- DUT-Omron Image dataset
- THUS
- ECSD
- PASCAL
- ...
More challenging situations with more noise
- Detection on magnetism such as MR: Medical image dataset and Extract voice from magnetic tape