Research map / Detection and segmentation / 3D detection and driving
Beyond Appearance: Camouflaged Object Detection via Geometric Structure
Jinyu Han, Changguang Wu, Fuming Sun, Jinhui Tang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Sparse Mixture-of-Experts Adapter (SMEA) to frozen MDE encoder (ablation) | DepthSAM-L · COD10K | weighted F-measure: 0.719 -> 0.867 | Improved |
| use sparse multi-expert SMEA instead of a standard adapter (ablation) | DepthSAM-L · COD10K | weighted F-measure: 0.840 -> 0.867 | Improved |
| increase activated experts from Top-K=2 to Top-K=4 (ablation) | DepthSAM-L · COD10K | weighted F-measure: 0.867 -> 0.869 | No clear effect |
| add Geometric-Semantic Fusion Module (GSFM) instead of simple concatenation (ablation) | DepthSAM-L · COD10K | weighted F-measure: 0.766 -> 0.867 | Improved |
| use GSFM instead of a standard FFN for fusion (ablation) | DepthSAM-L · COD10K | weighted F-measure: 0.829 -> 0.867 | Improved |
| use spatial-frequency SFRM instead of standard MHSA in GSFM (ablation) | DepthSAM-L · COD10K | weighted F-measure: 0.809 -> 0.867 | Improved |
| use Depth Anything v2 backbone instead of SAM with the same components (ablation) | DepthSAM-L · COD10K | S-measure: 0.880 -> 0.920 | Improved |
| increase input resolution from 384 to 512 (ablation) | DepthSAM-L · COD10K | S-measure: 0.905 -> 0.920 | Improved |
| apply COD-trained model zero-shot to video camouflaged detection | DepthSAM-L · MoCA-Mask | S-measure: 73.4 -> 78.4 vs ZoomNeXt | Improved |
| retrain DepthSAM-L on RGB salient object detection | DepthSAM-L · DUTS | S-measure: 0.932 -> 0.947 vs Samba | Improved |
| retrain DepthSAM-L on optical remote sensing SOD | DepthSAM-L · EORSSD | MAE: S 0.9449 -> 0.9503; MAE 0.0049 (HFCNet) vs 0.0053 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ZoomNeXt — ZoomNeXt: A Unified Collaborative Pyramid Network for Camouflaged Object Detection (TPAMI 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VCP-DCN: Beyond Visual Concealed Property via Depth Collaborative Network for Camouflaged Object Detection (ECCV 2026 · similarity 0.89)
- Rethinking Detecting Salient and Camouflaged Objects in Unconstrained Scenes (ICCV 2025 · similarity 0.88)
- Camouflage Anything: Learning to Hide using Controlled Out-painting and Representation Engineering (CVPR 2025 · similarity 0.87)
- Shift the Lens: Environment-Aware Unsupervised Camouflaged Object Detection (CVPR 2025 · similarity 0.87)
- Segment, Magnify and Reiterate: Detecting Camouflaged Objects the Hard Way (CVPR 2022 · similarity 0.87)
- MMCSBench: A Fine-Grained Benchmark for Large Vision-Language Models in Camouflage Scenes (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
