Modality-Specific Hierarchical Enhancement for RGB-D Camouflaged Object Detection

Yuzhen Niu; Yangqing Wang; Ri Cheng; Fusheng Li; Rongshen Wang; Zhichen Yang

cs.CVApr 3, 2026

Modality-Specific Hierarchical Enhancement for RGB-D Camouflaged Object Detection

Authors: Yuzhen Niu, Yangqing Wang, Ri Cheng, Fusheng Li, Rongshen Wang, Zhichen Yang

Organizations: College of Computer and Data Science, Fuzhou University, Fuzhou, China

Abstract

Camouflaged object detection (COD) is challenging due to high target-background similarity, and recent methods address this by complementarily using RGB-D texture and geometry cues. However, RGB-D COD methods still underutilize modality-specific cues, which limits fusion quality. We believe this is because RGB and depth features are fused directly after backbone extraction without modality-specific enhancement. To address this limitation, we propose MHENet, an RGB-D COD framework that performs modality-specific hierarchical enhancement and adaptive fusion of RGB and depth features. Specifically, we introduce a Texture Hierarchical Enhancement Module (THEM) to amplify subtle texture variations by extracting high-frequency information and a Geometry Hierarchical Enhancement Module (GHEM) to enhance geometric structures via learnable gradient extraction, while preserving cross-scale semantic consistency. Finally, an Adaptive Dynamic Fusion Module (ADFM) adaptively fuses the enhanced texture and geometry features with spatially varying weights. Experiments on four benchmarks demonstrate that MHENet surpasses 16 state-of-the-art methods qualitatively and quantitatively. Code is available at https://github.com/afdsgh/MHENet.

Figures & tables

Fig. 1: Texture enhancement enriches the texture details of the limbs (red boxes) and geometry enhancement strengthens the bat and its boundary activations (green boxes), enabling fusion to better combine complementary cues. RGB activates more on texture-rich limbs but less on the camouflaged bat, while depth complements the bat for complementary fusion.

Fig. 2: The overall architecture of the proposed MHENet, which consists of three key components, Texture Hierarchical Enhancement Module (THEM), Geometry Hierarchical Enhancement Module (GHEM), and Adaptive Dynamic Fusion Module (ADFM).

RGB-based COD Methods
Methods	Pub.Year	CHAMELEON				CAMO-Test				COD10K-Test				NC4K
Methods	Pub.Year	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$
SINet [ 7 ]	CVPR.20	0.872	0.936	0.806	0.034	0.745	0.804	0.644	0.092	0.776	0.864	0.631	0.043	0.808	0.871	0.723	0.058
BGNet [ 9 ]	IJCAI.22	0.901	0.943	0.850	0.027	0.812	0.857	0.749	0.073	0.831	0.901	0.722	0.033	0.851	0.907	0.788	0.044
ZoomNet [ 10 ]	CVPR.22	0.902	0.943	0.845	0.023	0.820	0.877	0.752	0.066	0.838	0.888	0.729	0.029	0.853	0.896	0.784	0.043
FSPNet [ 25 ]	CVPR.23	0.908	0.943	0.851	0.023	0.856	0.899	0.799	0.050	0.851	0.895	0.735	0.026	0.879	0.915	0.816	0.035

TABLE I: Comparison of our proposed method with 16 state-of-the-art methods on four benchmark datasets across four metrics:

S_{\alpha}\uparrow

E_{\varphi}\uparrow

F_{\beta}^{\omega}\uparrow

M\downarrow

. The best two results are highlighted in red and blue .

Fig. 3: Overview of the Adaptive Dynamic Fusion Module.

Fig. 4: Visual comparisons of some recent COD methods and ours on different types of samples. More comparisons are provided in the supplementary material. Best viewed by zooming in for more details.

NO.	Modules				COD10K-Test				NC4K-Test
NO.	B.	THEM	GHEM	ADFM	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$
1	$\checkmark$				0.873	0.925	0.786	0.022	0.893	0.930	0.842	0.032
2	$\checkmark$	$\checkmark$			0.883	0.934	0.806	0.020	0.895	0.932	0.849	0.031
3	$\checkmark$			$\checkmark$	0.885	0.938	0.813	0.020	0.898	0.937	0.856	0.030
4	$\checkmark$	$\checkmark$	$\checkmark$		0.887	0.938	0.813	0.020	0.900	0.937	0.855	0.030
5	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	0.889	0.942	0.817	0.019	0.902	0.939	0.859	0.029

TABLE II: Ablation on the THEM, GHEM, and ADFM. B. denotes baseline.

NO.	Inputs	COD10K-Test				NC4K-Test
NO.	Inputs	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$
6	w/o Semantic Block	0.885	0.939	0.816	0.020	0.898	0.935	0.856	0.030
7	w/o Texture Block	0.886	0.940	0.814	0.020	0.899	0.937	0.855	0.030
8	w/o Geometry Block	0.887	0.941	0.815	0.020	0.899	0.937	0.856	0.030
9	MHENet (Ours)	0.889	0.942	0.817	0.019	0.902	0.939	0.859	0.029

TABLE III: Ablation Study of the Key Block in THEM and GHEM.

NO.	Inputs	COD10K-Test				NC4K-Test
NO.	Inputs	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$
10	only Depth	0.844	0.902	0.740	0.028	0.869	0.912	0.807	0.039
11	only RGB	0.880	0.937	0.803	0.021	0.895	0.936	0.851	0.029
12	RGB-D	0.889	0.942	0.817	0.019	0.902	0.939	0.859	0.029

TABLE IV: Ablation Study of the Inputs of MHENet.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Fig. S1: Failure cases and potential extensions of MHENet under occlusion, ambiguous boundaries, and noisy depth scenarios.

Panel of Figure 10 — Fig. S2: Visualization of the features of our MHENet. Best viewed by zooming in for more details.

Panel of Figure 11 — Fig. S3: Visual comparisons of some recent COD methods and ours on different types of samples. Please zoom in for more details.

Depth Estimation Model	COD10K-Test				NC4K-Test
Depth Estimation Model	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$	$S_{\alpha}\uparrow$	$E_{\varphi}\uparrow$	$F_{\beta}^{\omega}\uparrow$	$M\downarrow$
Depth Anything V1	0.886	0.939	0.812	0.020	0.900	0.937	0.857	0.030
Depth Anything V2	0.894	0.941	0.825	0.019	0.905	0.941	0.865	0.029
Distill Any Depth	0.889	0.942	0.817	0.019	0.902	0.939	0.859	0.029

Appendix

TABLE S1: Performance comparison using different depth estimation models.

Methods/Metrics	SINet	BGNet	FSPNet	RISNet	Ours
Params (M) $\downarrow$	48.95	79.85	273.80	26.58	53.36
FLOPs (G) $\downarrow$	38.76	59.45	283.31	96.57	37.97
Speed (FPS) $\uparrow$	46.06	39.49	26.23	24.82	23.43

Appendix

TABLE S2: Complexity analysis of different methods.

Explore similar work

VCP-DCN: Beyond Visual Concealed Property via Depth Collaborative Network for Camouflaged Object Detection
Jul 30, 2026Songsong Duan, Xi Yang, Nannan WangCross-Modal Alignment Multimodal Fusion
the State Key Laboratory of Integrated Services Networks, School of Telecommunications Engineering, Xidian University, Xi’an, China
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
Modality-Agnostic Prompt Learning for Multi-Modal Camouflaged Object Detection
Apr 14, 2026Hao Wang, Jiqing Zhang, Xin Yang +4Camouflaged Object Detection Multimodal Learning
Information Science and Technology College, Dalian Maritime University, Liaoning, 116026, China · Dalian University of Technology, Liaoning, 116024, China · Beijing University of Technology, Beijing, 100124, China

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

VCP-DCN: Beyond Visual Concealed Property via Depth Collaborative Network for Camouflaged Object Detection

Songsong Duan, Xi Yang, Nannan Wang

the State Key Laboratory of Integrated Services Networks, School of Telecommunications Engineering, Xidian University, Xi’an, China

Cross-Modal Alignment Multimodal Fusion Camouflaged Object Detection

Open paper AskPDF arXiv

Camouflaged Object Detection (COD) aims to identify and segment camouflaged objects in complex environments, which are often concealed because their color and texture are similar to the background. Several existing COD methods introduce depth maps to boost detection performance via learning complementary RGB-D features, ignoring modality-specific characteristics of concealed objects in the depth domain. To address this issue, we propose a depth collaborative network, called VCP-DCN, to mine distinguishable multi-modality features beyond visual concealed prototype in depth domain. Specifically, VCP-DCN progressively performs multi-modality alignment, interaction, and fusion for the COD task. In the alignment stage, we propose a Separable Prototype Embedding (SPE) module to learn modality-consistency and modality-specific RGB/depth prototype tokens through prototype contrastive learning. Furthermore, we develop a Multi-modality Dual Attention (MDA) module to enhance the cross-modal feature representation through local response maps between modality-consistency RGB/depth prototype tokens and visual tokens on the interaction stage. Finally, we design a Depth Adaptive Injection (DAI) module to adaptively measure contribution of RGB/depth features with a decision-making mechanism, which calculates similarity distance between RGB/depth modality-specific prototype tokens and modality-consistency ones on the fusion stage. Extensive experiments demonstrate the effectiveness of our VCP-DCN on three authoritative datasets.

Abstract

Explore similar work

VCP-DCN: Beyond Visual Concealed Property via Depth Collaborative Network for Camouflaged Object Detection

Language-Guided Structure-Aware Network for Camouflaged Object Detection

Modality-Agnostic Prompt Learning for Multi-Modal Camouflaged Object Detection