Research map / Detection and segmentation / CNN features
OSFormer: One-Stage Camouflaged Instance Segmentation with Transformers
Jialun Pei, Tianyang Cheng, Deng-Ping Fan, He Tang, Chuanbo Chen, Luc Van Gool
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 LST encoder (ablation) | OSFormer · COD10K-Test | AP: 33.7 -> 41.0 | Improved |
| add location-guided queries instead of learnable embeddings (ablation) | OSFormer · COD10K-Test | AP: 35.0 -> 41.0 | Improved |
| use location-guided queries instead of zero-initialized queries (ablation) | OSFormer · COD10K-Test | AP: 34.7 -> 41.0 | Improved |
| add BC-FFN in place of vanilla FFN (ablation) | OSFormer · COD10K-Test | AP: 37.2 -> 41.0 | Improved |
| add CFF (multi-scale) instead of single T2 (ablation) | OSFormer · COD10K-Test | AP: 38.0 -> 41.0 | Improved |
| add reverse edge attention (ablation) | OSFormer · COD10K-Test | AP: 39.3 -> 41.0 | Improved |
| encoder layers 6 vs 1 (decoder 3) (ablation) | LST · COD10K-Test | AP: 37.0 -> 41.0 | Improved |
| increase decoder layers from 3 to 6 (encoder 6) (ablation) | LST · COD10K-Test | AP: 41.0 -> 40.6 | Hurt |
| add C2 to multi-scale inputs (C3-C5 to C2-C5) (ablation) | LST · COD10K-Test | AP: 41.0 -> 39.9 | Hurt |
| replace T2-T5 with CNN C2-C5 in CFF (ablation) | CFF · COD10K-Test | AP: 41.0 -> 35.4 | Hurt |
| swap ResNet-50 for Swin-T (ablation) | OSFormer · COD10K-Test | AP: 41.0 -> 47.7 | Improved |
| swap ResNet-50 for ResNet-101 (ablation) | OSFormer · COD10K-Test | AP: 41.0 -> 42.0 | Improved |
Later papers that compared with this method
2 later accepted papers report OSFormer as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Catch Me If You Can Describe Me: Open-Vocabulary Camouflaged Instance Segmentation with Diffusion (IJCV 2026 · better on 1 of 1 shared result)
- Camouflaged Instance Segmentation via Explicit De-Camouflaging (CVPR 2023 · better on 2 of 3 shared results)
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.
- SOLOv2 — SOLOv2: Dynamic and Fast Instance Segmentation (NeurIPS 2020) · better on 5 of 5 shared results
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
- Mask-Attention-Free Transformer for 3D Instance Segmentation (ICCV 2023 · similarity 0.86)
- PolarMask: Single Shot Instance Segmentation With Polar Representation (CVPR 2020 · similarity 0.86)
- Camouflaged Instance Segmentation via Explicit De-Camouflaging (CVPR 2023 · similarity 0.86)
- A Simple Single-Scale Vision Transformer for Object Detection and Instance Segmentation (ECCV 2022 · similarity 0.85)
- Instance Segmentation With Mask-Supervised Polygonal Boundary Transformers (CVPR 2022 · similarity 0.85)
- SpaCeFormer: Fast Proposal-Free Open-Vocabulary 3D Instance Segmentation (ICML 2026 · similarity 0.85)
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.
