.png){kind=icon}
.png){kind=icon}
Return
EUNet: evidential multi-evidence fusion for uncertainty-aware camouflaged-object detection
.png){kind=icon}
PRE
AI
L
J
H
DOI：10.1007/s00371-026-04690-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Camouflaged-object detection (COD) remains challenging because high foreground–background similarity, ambiguous boundaries, and partial occlusion often lead to unreliable predictions. To address this issue, we characterize camouflage-induced visual ambiguity as camouflaged uncertainty, including appearance uncertainty and occlusion uncertainty, and propose EUNet, an uncertainty-aware COD framework based on evidential deep learning. EUNet generates multi-level evidential predictions and employs a multi-evidence weighted fusion module to aggregate them. Specifically, belief Jensen–Shannon divergence is used to quantify disagreement among subjective opinions and assign consistency-aware weights, thereby suppressing conflicting evidence while preserving meaningful uncertainty. We further construct CUCOD, an uncertainty-oriented benchmark dataset designed to evaluate detection and uncertainty quality under different types of camouflage-induced ambiguity. Experiments on four public COD datasets demonstrate competitive detection performance, while quantitative uncertainty evaluation and category-wise analysis on CUCOD show that the estimated uncertainty is closely associated with prediction errors, ambiguous boundaries, and occluded regions. The source code and datasets are available at: https://github.com/huangliucheng/EUNet .
Keywords:
Camouflaged-object detection
Uncertainty quantification
Evidential deep learning
Multi-evidence fusion
Journal
IF:
2.9
Papers:
4.5K
Citations:
6.5K
Organization
Cited Papers