CAMotion: A High-Quality Benchmark for Camouflaged Moving Object Detection in the Wild

Siyuan Yao; Hao Sun; Ruiqi Yu; Xiwei Jiang; Wenqi Ren; Xiaochun Cao

cs.CVApr 9, 2026

CAMotion: A High-Quality Benchmark for Camouflaged Moving Object Detection in the Wild

Authors: Siyuan Yao, Hao Sun, Ruiqi Yu, Xiwei Jiang, Wenqi Ren, Xiaochun Cao

Organizations: School of CyberScience and Technology, Sun Yat-sen University, Shenzhen Campus, Shenzhen 518107, China. · the College of Computing and Data Science, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798. · School of Computer Science (National Pilot Software Engineering School), Beijing University of Posts and Telecommunications, Beijing 100876, China.

Abstract

Discovering camouflaged objects is a challenging task in computer vision due to the high similarity between camouflaged objects and their surroundings. While the problem of camouflaged object detection over sequential video frames has received increasing attention, the scale and diversity of existing video camouflaged object detection (VCOD) datasets are greatly limited, which hinders the deeper analysis and broader evaluation of recent deep learning-based algorithms with data-hungry training strategy. To break this bottleneck, in this paper, we construct CAMotion, a high-quality benchmark covers a wide range of species for camouflaged moving object detection in the wild. CAMotion comprises various sequences with multiple challenging attributes such as uncertain edge, occlusion, motion blur, and shape complexity, etc. The sequence annotation details and statistical distribution are presented from various perspectives, allowing CAMotion to provide in-depth analyses on the camouflaged object's motion characteristics in different challenging scenarios. Additionally, we conduct a comprehensive evaluation of existing SOTA models on CAMotion, and discuss the major challenges in VCOD task. The benchmark is available at https://www.camotion.focuslab.net.cn, we hope that our CAMotion can lead to further advancements in the research community.

Figures & tables

Fig. 1: Examples of our CAMotion dataset with corresponding pixel-level annotations. The first and third rows contain original images; the second and last rows contain corresponding pixel-wise ground truth annotations.

Dataset	Year	Publication	Type	#Img.	#Ann. Img.	#Species	Bbox. GT	Mask GT
CAD [ 5 ]	2016	ECCV	Video	839	181	6	✗	✓
CHAMELEON [ 2 ]	2018	-	Image	76	76	27 ∗	✗	✓
CAMO [ 1 ]	2019	CVIU	Image	2,500	1,250	97 ∗	✗	✓
COD10K [ 3 ]	2020	CVPR	Image	10,000	5,066	69	✓	✓
MoCA [ 7 ]	2020	ACCV	Video	37,250	7,617	67	✓	✗
CAMO++ [ 8 ]	2021	TIP	Image	5,500	5,500	93	✓	✓

TABLE I: Statistics of camouflage datasets. ∗ indicates that the #Species is not reported in the original paper and is counted by us.

Fig. 2: Dataset features and species examples from CAMotion dataset. (a) Taxonomic structure of CAMotion. (b) The scale and species comparison between existing COD dataset and CAMotion. (c) The attributes distribution in frame-level and sequence-level. (d) Examples of the Ray-finned Fish class in CAMotion. Please zoom in for details.

Fig. 3: Visualization of the challenging attributes in CAMotion. Best viewed in color and zoomed in for details.

Fig. 4: Examples of refined initial annotations. White denotes unchanged regions; red and green indicate over-annotated and previously missing regions in the original annotations, respectively. Please zoom in for details.

Attr	Description
MO	Multiple Objects : image contains at least two objects.
BO	Big Object : ratio between object area and image area $\geq$ 0.15.
SO	Small Object : ratio between object area and image area $\leq$ 0.02.
UE	Uncertain Edge : the foreground and background areas around object have similar colors and textures.
OC	Occlusion : the object is partially occluded.
SC	Shape Complexity : object contains thin parts (e.g., animal foot).

TABLE II: List and description of the eight attributes that characterize videos in CAMotion.

Fig. 5: Statistics for CAMotion dataset. (a) Object sizes distribution. (b) The distribution of video durations. (c) Global/local contrast distribution. (d) Motion statistics of the camouflaged objects.

Methods	Publications	CAMotion						MoCA-Mask
Methods	Publications	$S_{\alpha}\uparrow$	$F_{\beta}^{w}\uparrow$	$E_{\phi}^{m}\uparrow$	$\mathcal{M}\downarrow$	$\mathrm{mDic}\uparrow$	$\mathrm{mIoU}\uparrow$	$S_{\alpha}\uparrow$	$F_{\beta}^{w}\uparrow$	$E_{\phi}^{m}\uparrow$	$\mathcal{M}\downarrow$	$\mathrm{mDic}\uparrow$	$\mathrm{mIoU}\uparrow$
MGL-R [ 15 ]	CVPR’21	0.542	0.176	0.604	0.078	0.195	0.129	0.493	0.034	0.519	0.059	0.048	0.033
PFNet [ 80 ]	CVPR’21	0.669	0.425	0.780	0.050	0.463	0.359	0.558	0.142	0.633	0.026	0.172	0.118
UGTR [ 81 ]	ICCV’21	0.687	0.403	0.720	0.048	0.440	0.342	0.493	0.048	0.459	0.088	0.078	0.049
SegMaR [ 76 ]	CVPR’22	0.645	0.377	0.695	0.049	0.402	0.311	0.542	0.129	0.544	0.024	0.139	0.093
ZoomNet [ 13 ]	CVPR’22	0.675	0.440	0.693	0.044	0.456	0.365	0.582	0.201	0.682	0.026	0.236	0.197

TABLE III: Quantitative comparison with 18 cutting-edge methods on CAMotion and MoCA-Mask testing datasets. Notes

\uparrow/\downarrow

denotes the higher/lower the better, and the best and second best are bolded and underlined for highlighting, respectively. ‡ indicates that the first-frame annotation is removed during both training and testing for fair comparison.

Panel of Table 9 — TABLE IV: $S_{\alpha}$ and $\mathcal{M}$ results for cross-dataset generalization. The selected ZoomNeXt is trained on one (rows) dataset and tested on all datasets (columns). “Self” refers to training and testing on the same dataset (same as diagonal), and “Mean Others” refers to averaging performance on all except self.

Fig. 6: Visual comparison with state-of-the-art methods in challenging scenarios, i.e. , shape complexity (Rows 1-4) and occlusion (Rows 5-8). Please zoom in for details.

Fig. 7: Optical flow and depth properties visualization. Each group comprises the original image, pixel-level annotation, optical flow, and depth map. Please zoom in for details.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Fig. 12: Detailed classification of the total frame count across different species. Please zoom in for details.

Fig. 13: Examples of our CAMotion dataset with corresponding pixel-level annotations, optical flow and depth map. Each group is arranged as the original image, optical flow, depth map and pixel-level annotation. Please zoom in for details.

Fig. 14: Visualization of the challenging attributes in CAMotion. Best viewed in color and zoom in for details.

Fig. 15: Examples of fine-tuning initial annotations. The white color denotes unchanged areas, while the red and green indicate the original and refined annotations, respectively. Please zoom in for details.

Fig. 16: Visualization of the bounding box annotations in CAMotion. Best viewed in color and zoom in for details.

Fig. 17: Visual comparison with state-of-the-art methods on CAMotion. Please zoom in for details.

Fig. 20: Failure cases on both HGINet and ZoomNeXt. Please zoom in for details.

Explore similar work

YUV20K: A Complexity-Driven Benchmark and Trajectory-Aware Alignment Model for Video Camouflaged Object Detection
Apr 11, 2026Yiyu Liu, Shuo Ye, Chao Hao +1Camouflaged Object Detection
Yiyu Liu, Shuo Ye, Chao Hao and Zitong Yu are with the Department of Computer Science, Great Bay University, Dongguan, Guangdong, 523000, China
COD10K-C: Benchmarking Robustness of Camouflaged Object Detection Under Natural Image Corruptions
May 23, 2026Arafat Hossain SayemImage Segmentation Image Corruption Robustness
Is There Really a Camouflaged Object? Towards Realistic Camouflaged Object Detection
Aug 11, 2026Huafeng Chen, Yueming Lyu, Chenyang Si +3Object Detection Camouflaged Object Detection
School of Intelligence Science and Technology, Nanjing University · Department of Computing, Imperial College London

Apr 11, 2026J/K move · Enter open · S save

YUV20K: A Complexity-Driven Benchmark and Trajectory-Aware Alignment Model for Video Camouflaged Object Detection

Yiyu Liu, Shuo Ye, Chao Hao, Zitong Yu

Yiyu Liu, Shuo Ye, Chao Hao and Zitong Yu are with the Department of Computer Science, Great Bay University, Dongguan, Guangdong, 523000, China

Camouflaged Object Detection

Open paper AskPDF arXiv

Video Camouflaged Object Detection (VCOD) is currently constrained by the scarcity of challenging benchmarks and the limited robustness of models against erratic motion dynamics. Existing methods often struggle with Motion-Induced Appearance Instability and Temporal Feature Misalignment caused by complex motion scenarios. To address the data bottleneck, we present YUV20K, a pixel-level annoated complexity-driven VCOD benchmark. Comprising 24,295 annotated frames across 91 scenes and 47 kinds of species, it specifically targets challenging scenarios like large-displacement motion, camera motion and other 4 types scenarios. On the methodological front, we propose a novel framework featuring two key modules: Motion Feature Stabilization (MFS) and Trajectory-Aware Alignment (TAA). The MFS module utilizes frame-agnostic Semantic Basis Primitives to stablize features, while the TAA module leverages trajectory-guided deformable sampling to ensure precise temporal alignment. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art competitors on existing datasets and establishes a new baseline on the challenging YUV20K. Notably, our framework exhibits superior cross-domain generalization and robustness when confronting complex spatiotemporal scenarios. Our code and dataset will be available at https://github.com/K1NSA/YUV20K

Abstract

Explore similar work

YUV20K: A Complexity-Driven Benchmark and Trajectory-Aware Alignment Model for Video Camouflaged Object Detection

COD10K-C: Benchmarking Robustness of Camouflaged Object Detection Under Natural Image Corruptions

Is There Really a Camouflaged Object? Towards Realistic Camouflaged Object Detection