A co-segmentation dataset characterized by simple background(MLMR-COS)
Published: 6 September 2018| Version 1 | DOI: 10.17632/xbxrc6d3xw.1
Contributor:
Yu MengqiaoDescription
The new proposed dataset for evaluating co-segmentation algorithms, which contains 66 image groups with 18 images in each and the pixel-wise hand-annotated ground truth. The dataset is characterized by simple background produced from a single color. It looks simple but actually very challenging for current co-segmentation algorithms, because of four difficult cases in it: easy confused foreground with background, transparent regions in objects, minor holes in objects, and shadows.
Files
Categories
Image Segmentation, Database