EviRCOD: Evidence-Guided Probabilistic Decoding for Referring Camouflaged Object Detection

Ye Wang; Kai Huang; Sumin Shen; Chenyang Ma

cs.CVApr 13, 2026

EviRCOD: Evidence-Guided Probabilistic Decoding for Referring Camouflaged Object Detection

Authors: Ye Wang, Kai Huang, Sumin Shen, Chenyang Ma

Organizations: West China Hospital, Sichuan University · Independent Researcher · Department of Data Science and Artificial Intelligence, Auckland University of Technology

Abstract

Referring Camouflaged Object Detection (Ref-COD) focuses on segmenting specific camouflaged targets in a query image using category-aligned references. Despite recent advances, existing methods struggle with reference-target semantic alignment, explicit uncertainty modeling, and robust boundary preservation. To address these issues, we propose EviRCOD, an integrated framework consisting of three core components: (1) a Reference-Guided Deformable Encoder (RGDE) that employs hierarchical reference-driven modulation and multi-scale deformable aggregation to inject semantic priors and align cross-scale representations; (2) an Uncertainty-Aware Evidential Decoder (UAED) that incorporates Dirichlet evidence estimation into hierarchical decoding to model uncertainty and propagate confidence across scales; and (3) a Boundary-Aware Refinement Module (BARM) that selectively enhances ambiguous boundaries by exploiting low-level edge cues and prediction confidence. Experiments on the Ref-COD benchmark demonstrate that EviRCOD achieves state-of-the-art detection performance while providing well-calibrated uncertainty estimates. Code is available at: https://github.com/blueecoffee/EviRCOD.

Figures & tables

Fig. 1: Visual comparisons of different methods. The proposed EviRCOD generates more complete segmentation structures, sharper boundaries, and fewer background misclassifications compared to R2CNet [ 2 ] and UAT [ 3 ] .

Fig. 2: Architecture of EviRCOD, illustrating the Reference-Guided Deformable Encoder, the Uncertainty-Aware Evidential Decoder, and the Boundary-Aware Refinement Module for uncertainty-aware Referring Camouflaged Object Detection.

Fig. 3: Qualitative comparisons between EviRCOD and the main Ref-COD baselines (R2CNet [ 2 ] and UAT [ 3 ] ). EviRCOD produces more coherent object masks with sharper boundaries and significantly fewer background errors, highlighting its superiority in challenging camouflage scenarios.

Models	Pub&Year	Overall				Single Object				Multiple Objects
		$S_{m}\uparrow$	$\alpha_{E}\uparrow$	$\omega_{F}\uparrow$	$\mathcal{M}\downarrow$	$S_{m}\uparrow$	$\alpha_{E}\uparrow$	$\omega_{F}\uparrow$	$\mathcal{M}\downarrow$	$S_{m}\uparrow$	$\alpha_{E}\uparrow$	$\omega_{F}\uparrow$	$\mathcal{M}\downarrow$
PFNet-Ref	PFNet [ 8 ] : CVPR-2021	0.811	0.885	0.687	0.036	0.815	0.886	0.691	0.035	0.764	0.873	0.632	0.045
PreyNet-Ref	PreyNet [ 9 ] : ACMMM-2022	0.817	0.900	0.704	0.032	0.822	0.900	0.709	0.032	0.763	0.898	0.645	0.041
SINetV2-Ref	SINetV2 [ 10 ] : TPAMI-2022	0.823	0.888	0.700	0.033	0.828	0.889	0.705	0.032	0.771	0.874	0.634	0.043
BGNet-Ref	BGNet [ 11 ] : IJCAI-2022	0.840	0.909	0.738	0.029	0.844	0.910	0.742	0.029	0.792	0.887	0.679	0.036
BSANet-Ref	BSANet [ 12 ] : AAAI-2022	0.830	0.912	0.727	0.030	0.827	0.913	0.733	0.030	0.774	0.895	0.655	0.039

TABLE I: QUANTITATIVE COMPARISON OF EviRCOD WITH STATE-OF-THE-ART METHODS. The best three results are highlighted with bold (1st), solid underline (2nd), and dashed underline (3rd). “

\uparrow

”: the higher the better, “

\downarrow

”: the lower the better.

Configuration	$S_{m}\uparrow$	$\alpha_{E}\uparrow$	$\omega_{F}\uparrow$	$\mathcal{M}\downarrow$
Baseline	0.838	0.908	0.751	0.027
Baseline + RGDE	0.854	0.928	0.775	0.023
Baseline + UAED	0.859	0.932	0.780	0.024
Baseline + BARM	0.848	0.917	0.760	0.025
Baseline + RGDE + UAED	0.865	0.940	0.794	0.022
Baseline + RGDE + BARM	0.860	0.933	0.781	0.022

TABLE II: Ablation on R2C7K: module contributions to structure, alignment, contour, and pixel-wise metrics.

Configuration	$S_{m}\uparrow$	$\alpha_{E}\uparrow$	$\omega_{F}\uparrow$	$\mathcal{M}\downarrow$
EUQM	0.867	0.940	0.793	0.021
EGA	0.864	0.937	0.790	0.021
UMRM	0.865	0.938	0.791	0.021
EUQM+EGA	0.867	0.941	0.796	0.021
EUQM+UMRM	0.868	0.942	0.797	0.021
EGA+UMRM	0.865	0.939	0.792	0.021

TABLE III: Ablation of UAED sub-modules (EUQM, EGA, and UMRM).

Fig. 4: Calibration comparison between UAT and our method.

Explore similar work

Consensus-Aware Multi-Source Fusion for Reference-Guided Camouflaged Object Detection
Sep 30, 2026Junyang Xia, Luocheng Zhang, Wenwen Pan +4Multi-Scale Feature Fusion Camouflaged Object Detection
School of Computer Science, Hangzhou Dianzi University, Hangzhou 310018, China
EReCu: Pseudo-label Evolution Fusion and Refinement with Multi-Cue Learning for Unsupervised Camouflage Detection
Mar 12, 2026Shuo Jiang, Gaojia Zhang, Min Tan +2Pseudo-Labeling Camouflaged Object Detection
Zhejiang Key Laboratory of Space Information Sensing and Transmission, Hangzhou Dianzi University · Laboratory of Complex Systems Modeling and Simulation, School of Computer Science and Technology, Hangzhou Dianzi University · College of Computer Science and Technology, Zhejiang University
Language-Guided Structure-Aware Network for Camouflaged Object Detection
Mar 25, 2026Min ZhangCamouflaged Object Detection Boundary-Aware Segmentation
School of Artificial Intelligence, Chongqing University of Technology, Chongqing, 401120, China

Sep 30, 2026J/K move · Enter open · S save

Consensus-Aware Multi-Source Fusion for Reference-Guided Camouflaged Object Detection

Junyang Xia, Luocheng Zhang, Wenwen Pan, Chifeng Zhu+3

School of Computer Science, Hangzhou Dianzi University, Hangzhou 310018, China

Multi-Scale Feature Fusion Camouflaged Object Detection

Open paper AskPDF arXiv

Reference-guided camouflaged object detection aims to segment a target whose visual appearance closely resembles its surroundings by exploiting auxiliary reference samples. The task remains difficult because reference samples contain inconsistent target cues, while generic visual representations are not inherently aligned with the target specified by the references. To handle these problems, we present a consensus-aware multi-source fusion framework. Reference-Conditioned Dual-Backbone Fusion (RCDF) couples trainable PVTv2 query features with frozen DINOv3 representations and uses reference-conditioned correlation to select foundation-model evidence before multi-scale fusion. The framework also aggregates multiple references through cross-reference consensus aggregation and injects reference information at semantic depths matched to the query features. Extensive experiments demonstrate the effectiveness of the proposed method. The results further show that reference consensus, target-conditioned foundation features, and hierarchical decoding provide complementary improvements under the evaluation protocol. The source code will be made publicly available upon acceptance.

Abstract

Explore similar work

Consensus-Aware Multi-Source Fusion for Reference-Guided Camouflaged Object Detection

EReCu: Pseudo-label Evolution Fusion and Refinement with Multi-Cue Learning for Unsupervised Camouflage Detection

Language-Guided Structure-Aware Network for Camouflaged Object Detection