Research map / Detection and segmentation / Open-set and few-shot detection
Catch Me If You Can Describe Me: Open-Vocabulary Camouflaged Instance Segmentation with Diffusion
Tuan-Anh Vu, Duc Thanh Nguyen, Qing Guo, Nhat Minh Chung, Binh‐Son Hua, Ivor Wai-Hung Tsang, Sai-Kit Yeung
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 |
|---|---|---|---|
| fine-tune on COD10K-v3 training set | Ours · COD10K-v3 test | AP: 23.9 -> 45.1 | Improved |
| fine-tune on COD10K-v3 training set | Ours · NC4K | AP: 24.8 -> 52.9 | Improved |
| add text embeddings (vs zero text embeddings) (ablation) | Ours · COD10K-v3 test | AP: -7.1 (19.3 vs 12.2) | Improved |
| add text embeddings (vs zero) after task-specific fine-tuning (ablation) | Ours (task-specific) · COD10K-v3 test | AP: 44.9 vs 31.4 (-13.5) | Improved |
| add MSFF multi-scale fusion over all U-Net layers (vs last layer only) (ablation) | Ours · COD10K-v3 test | AP: -0.9 (18.4 vs 19.3) | Improved |
| add MSFF fusion (vs concatenation of all multiscale features) (ablation) | Ours · COD10K-v3 test | AP: -1.2 (18.1 vs 19.3) | Improved |
| add CIN module (ablation) | Ours · COD10K-v3 test | AP: 19.3 -> from 17.6 without it | Improved |
| add TVA module (vs element-wise dot product) (ablation) | Ours · COD10K-v3 test | AP: -0.5 (18.8 vs 19.3) | Improved |
| add prompt engineering (synonyms, plurals ensemble) (ablation) | Ours · COD10K-v3 test | AP: 22.8 -> 23.4 | Improved |
| add multiple camouflage-specific prompt templates (ablation) | Ours · COD10K-v3 test | AP: 23.4 -> 23.9 | Improved |
| replace fine-grained class names with coarser hypernym prompts (ablation) | Ours · COD10K-v3 test | AP: 23.9 -> 22.7 | Hurt |
| replace fine-grained class names with coarser hypernym prompts (ablation) | Ours · NC4K | AP: 24.8 -> 23.6 | Hurt |
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.
- OSFormer — OSFormer: One-Stage Camouflaged Instance Segmentation with Transformers (ECCV 2022) · 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
- Open-Vocabulary Camouflaged Object Segmentation (ECCV 2024 · similarity 0.93)
- Training-Free Open-Vocabulary Camouflaged Object Segmentation via Fine-Grained Object Binding and Adaptive Hybrid Prompt (CVPR 2026 · similarity 0.91)
- Seeing the Unseen: A Semantic Alignment and Context-Aware Prompt Framework for Open-Vocabulary Camouflaged Object Segmentation (ICCV 2025 · similarity 0.90)
- Iterative Refinement of Semantic and Spatial Representations for Open-Vocabulary Camouflaged Object Segmentation (ECCV 2026 · similarity 0.90)
- Training-Free Open-Vocabulary Segmentation with Offline Diffusion-Augmented Prototype Generation (CVPR 2024 · similarity 0.88)
- Shift the Lens: Environment-Aware Unsupervised Camouflaged Object Detection (CVPR 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.
