mlsanityData to 7 Oct 2026

From Reconstruction to Decision: A Post-Encoder Plug-in Adapter for Curvilinear Segmentation

arXiv 2606.23486 ↗First posted 22 Jun 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Curvilinear object segmentation, including vessels and cracks, is challenging due to extreme spatial sparsity and topological fragility, where small local errors can cause severe structural disconnections. Meanwhile, modern segmentation pipelines increasingly rely on strong but hard-to-modify foundation encoders whose heavy downsampling limits fine structural recovery. Motivated by this, we focus on the post-encoder stage and study two recurring and actionable failure modes: a reconstruction bottleneck in high-resolution feature restoration and a decision bottleneck in binarization. We present PEPA, a lightweight Post-Encoder Plug-in Adapter for 2D curvilinear segmentation pipelines with accessible decoder/head features and target, query, or class descriptors. PEPA couples (i) Target-Conditioned Snake Upsampling (TCSU), which uses target-conditioned continuous snake-like sampling to better recover thin and tortuous structures during upsampling, and (ii) Target-Adaptive Differentiable Thresholding (TADT), which predicts target-specific thresholds and optimizes a soft-threshold surrogate with explicit safeguards against trivial bias shifting. Under this post-encoder interface, PEPA can be attached to both prompt-based decoders and conventional dense predictors. Experiments on five medical and industrial benchmarks show that adding PEPA to frozen-encoder baselines yields consistent improvements, with gains in topological connectivity (clDice) typically exceeding those in region overlap (IoU), indicating improved structural continuity. With only ~0.26M additional parameters, PEPA offers a practical post-encoder enhancement for structure-centric segmentation.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 32 of its 56 references that are in the map. First posted on arXiv 22 Jun 2026.

Builds on

32 of 56 references.

References · 32

  1. 2023Segment Anythingcited by 8,158
  2. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  3. 2024SAM 2: Segment Anything in Images and Videoscited by 2,784
  4. 2021Masked-attention Mask Transformer for Universal Image Segmentationcited by 2,338
  5. 2025DINOv3cited by 1,297
  6. 2023Segment Anything in High Qualitycited by 373
  7. 2022Mask DINO: Towards A Unified Transformer-based Framework for Object Detection and Segmentationcited by 341
  8. 2020clDice -- A Novel Topology-Preserving Loss Function for Tubular Structure Segmentationcited by 286
  9. 2019Real-time Scene Text Detection with Differentiable Binarizationcited by 212
  10. 2023EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment Anythingcited by 187
  11. 2019CARAFE: Content-Aware ReAssembly of FEaturescited by 119
  12. 2024FeatUp: A Model-Agnostic Framework for Features at Any Resolutioncited by 114
20 more
  1. 2022Real-Time Scene Text Detection with Differentiable Binarization and Adaptive Scale Fusioncited by 94
  2. 2023Dynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentationcited by 83
  3. 2019Feature Pyramid and Hierarchical Boosting Network for Pavement Crack Detectioncited by 80
  4. 2024SAM2-Adapter: Evaluating & Adapting Segment Anything 2 in Downstream Tasks: Camouflage, Shadow, Medical Image Segmentation, and Morecited by 58
  5. 2023Learning to Upsample by Learning to Samplecited by 57
  6. 2020CS2-Net: Deep Learning Segmentation of Curvilinear Structures in Medical Imagingcited by 52
  7. 2025AnyUp: Universal Feature Upsamplingcited by 26
  8. 2025LoftUp: Learning a Coordinate-Based Feature Upsampler for Vision Foundation Modelscited by 23
  9. 2025JAFAR: Jack up Any Feature at Any Resolutioncited by 21
  10. 2024Binarized Diffusion Model for Image Super-Resolutioncited by 18
  11. 2018Deep convolutional neural networks for segmenting 3D in vivo multiphoton images of vasculature in Alzheimer disease mouse modelscited by 12
  12. 2022DTU-Net: Learning Topological Similarity for Curvilinear Structure Segmentationcited by 12
  13. 2025BiMaCoSR: Binary One-Step Diffusion Model Leveraging Flexible Matrix Compression for Real Super-Resolutioncited by 5
  14. 2024Enriching Information and Preserving Semantic Consistency in Expanding Curvilinear Object Segmentation Datasetscited by 4
  15. 2025CrossDiff: Diffusion Probabilistic Model With Cross-conditional Encoder-Decoder for Crack Segmentationcited by 4
  16. 2025VesSAM: Efficient Multi-Prompting for Segmenting Complex Vesselcited by 3
  17. 2024Retinal Vessel Segmentation with Deep Graph and Capsule Reasoningcited by 2
  18. 2025CAPE: Connectivity-Aware Path Enforcement Loss for Curvilinear Structure Delineationcited by 2
  19. 2025UCS: A Universal Model for Curvilinear Structure Segmentationcited by 2
  20. 2025Binarization-Aware Adjuster for Discrete Decision Learning with an Application to Edge Detectioncited by 1

Built on by

Not cited yet.