Research map / Video / Video segmentation and tracking
The Making and Breaking of Camouflage
Hala Lamdouar, Weidi Xie, Andrew Zisserman
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 intra-image Frechet loss to the camouflage GAN generator (ablation) | StyleGAN camouflage generator · synthetic camouflage images | camouflage score S_alpha: 0.546 -> 0.598 | Improved |
| add intra-image Frechet loss to the GAN (ablation) | StyleGAN camouflage generator · synthetic camouflage images | FID: 75.01 -> 76.34 | Hurt |
| add intra-image Frechet loss to the GAN (ablation) | StyleGAN camouflage generator · synthetic camouflage images | Inception Score (object clarity): 3.49 -> 3.13 | Hurt |
| pre-train on synthetic data then fine-tune on MoCA-Mask instead of MoCA-Mask only (ablation) | full motion segmentation model · MoCA-Mask test | mIoU: 23.4 -> 30.8 | Improved |
| add synthetic data to MoCA-Mask training of the simple model (ablation) | baseline model without appearance encoder · MoCA-Mask test | mIoU: 15.3 -> 16.0 | Improved |
| add appearance encoder (ablation) | motion segmentation model · MoCA-Mask test | mIoU: 16.0 -> 20.5 | Improved |
| add transformer encoder on top of the appearance encoder (ablation) | motion segmentation model · MoCA-Mask test | mIoU: 20.5 -> 23.0 | Improved |
| add transformer aggregation to the model with transformer encoder (ablation) | motion segmentation model · MoCA-Mask test | mIoU: 23.0 -> 30.8 | Improved |
| train on synthetic data only (ablation) | baseline model · MoCA-Mask test | mIoU: 14.5 vs 15.3 (MM only) | Hurt |
| combine S_Rf and S_b (alpha=0.35) instead of either alone (ablation) | camouflage score · CHAMELEON human ranking | Kendall-tau: 0.01 / 0.03 -> 0.42 | Improved |
| use boundary score instead of reconstruction fidelity (ablation) | camouflage score · Camouflaged cuboids human ranking | Kendall-tau: -0.07 -> 0.42 | Improved |
| extend synthetic images to video sequences | synthetic camouflage generation · synthetic camouflage | camouflage score S_alpha: 0.598 -> 0.578 | Hurt |
Closest papers by meaning
- Unlocking Attributes' Contribution to Successful Camouflage: A Combined Textual and Visual Analysis Strategy (ECCV 2024 · similarity 0.89)
- Simultaneously Localize, Segment and Rank the Camouflaged Objects (CVPR 2021 · similarity 0.89)
- Camouflage Anything: Learning to Hide using Controlled Out-painting and Representation Engineering (CVPR 2025 · similarity 0.88)
- CamSAM2: Segment Anything Accurately in Camouflaged Videos (NeurIPS 2025 · similarity 0.85)
- Strategic Preys Make Acute Predators: Enhancing Camouflaged Object Detectors by Generating Camouflaged Objects (ICLR 2024 · similarity 0.84)
- CamoDiffusion: Camouflaged Object Detection via Conditional Diffusion Models (AAAI 2024 · similarity 0.84)
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.
