Research map / Detection and segmentation / CNN features
From Reconstruction to Decision: A Post-Encoder Plug-in Adapter for Curvilinear Segmentation
Qin Lei, Jiang Zhong, Xin Xiao, ymyang ymyang, Hao Wu
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 PEPA to frozen-encoder SAM | SAM · Avg of 5 datasets | IoU / clDice: 66.9 -> 69.9 IoU; 79.9 -> 83.1 clDice | Improved |
| add PEPA to EfficientSAM | EfficientSAM · Avg of 5 datasets | IoU / clDice: 64.2 -> 67.2 IoU; 76.8 -> 80.2 clDice | Improved |
| add PEPA to DINOv3 with Mask2Former | DINOv3 (Mask2Former) · Avg of 5 datasets | IoU / clDice: 65.7 -> 67.6 IoU; 79.2 -> 81.1 clDice | Improved |
| add PEPA to SAM | SAM · CHASEDB1 | clDice: 79.9 -> 85.5 (+5.6) | Improved |
| add TCSU to SAM (ablation) | SAM · XCAD | IoU / clDice: 68.4 -> 71.0 IoU; 81.5 -> 83.5 clDice | Improved |
| add TADT to SAM (ablation) | SAM · XCAD | IoU / clDice: 68.4 -> 70.2 IoU; 81.5 -> 84.0 clDice | Improved |
| add both TCSU and TADT to SAM (ablation) | SAM · XCAD | IoU / clDice: 68.4 -> 73.1 IoU; 81.5 -> 85.3 clDice | Improved |
| add both TCSU and TADT to nnU-Net (ablation) | nnU-Net · XCAD | IoU / clDice: 70.6 -> 72.4 IoU; 81.8 -> 83.8 clDice | Improved |
| add clDice loss on the soft-threshold surrogate in TADT (ablation) | PEPA SAM · XCAD | clDice: 82.8 -> 85.3 | Improved |
| add logit centering to TADT (ablation) | PEPA SAM · XCAD | clDice: 84.9 -> 85.3 | Improved |
| add consistency loss to TADT (ablation) | PEPA SAM · XCAD | clDice: 84.8 -> 85.3 | Improved |
| PEPA versus encoder-side LoRA with matched 0.26M parameters | SAM · XCAD and Crack500 average | clDice: 80.9 -> 82.5 | Improved |
Closest papers by meaning
- Dual-level Adapter Boosting Prompt-free Curvilinear Structure Segmentation (CVPR 2026 · similarity 0.88)
- Enriching Information and Preserving Semantic Consistency in Expanding Curvilinear Object Segmentation Datasets (ECCV 2024 · similarity 0.88)
- Joint Segmentation and Path Classification of Curvilinear Structures (TPAMI 2019 · similarity 0.88)
- Joint Topology-Preserving and Feature-Refinement Network for Curvilinear Structure Segmentation (ICCV 2021 · similarity 0.87)
- DeepBranchTracer: A Generally-Applicable Approach to Curvilinear Structure Reconstruction Using Multi-Feature Learning (AAAI 2024 · similarity 0.87)
- TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention (NeurIPS 2026 · 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.
