.png){kind=icon}
.png){kind=icon}
Return
Adaptive context mining for camouflaged object detection with scribble supervision
.png){kind=icon}
PRE
AI
DOI：10.1016/j.cviu.2025.104430.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Proposed SCNet, a novel weakly supervised COD framework using scribble annotations, reducing reliance on pixel-level labels. • Implemented a two-stage approach with NID for coarse localization and RM for precise refinement, mimicking human visual behavior. • Designed ALCC loss to adaptively adjust attention window size based on local complexity, improving model adaptability to complex scenarios. • Achieved state-of-the-art performance on COD benchmarks, surpassing weakly supervised methods and competing with fully supervised approaches. • Applied SCNet to polyp segmentation, achieving excellent performance comparable to fully supervised methods.
Keywords:
SCNet
weakly supervised COD
scribble annotations
NID
RM
ALCC loss
Journal
IF:
3.5
Papers:
441
Citations:
7.3K
Organization
No organization information available