Research map / Detection and segmentation / Open-set and few-shot detection
Enriching Information and Preserving Semantic Consistency in Expanding Curvilinear Object Segmentation Datasets
Qin Lei, Jiang Zhong, Qizhu Dai
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 |
|---|---|---|---|
| replace GroupNorm with SPADE in ControlNet encoder blocks (SCP ControlNet) (ablation) | ControlNet for semantic image synthesis · Crack500 5x expanded, UNet | mIoU: 75.4 -> 78.4 | Improved |
| add SPADE in Down.3 encoder block (on top of Down.1-2) (ablation) | SCP ControlNet · Crack500 5x expanded, UNet | mIoU: 75.9 -> 77.9 | Improved |
| add SPADE in Middle block on top of Down.1-4 (ablation) | SCP ControlNet · Crack500 5x expanded, UNet | mIoU: 78.2 -> 78.4 | Improved |
| apply SPADE only in the early encoder blocks (Down.1 or Down.2) (ablation) | SCP ControlNet · Crack500 5x expanded, UNet | mIoU: 75.4 -> 75.7 (Down.1), 75.5 (Down.2) | No clear effect |
| expand dataset 5x with SCP ControlNet synthetic data vs training on original | UNet trained from scratch · CHUAC | mIoU: 51.2 -> 61.5 | Improved |
| expand with RandAugment 5x | UNet on DRIVE · DRIVE | mIoU: 68.6 -> 67.2 | Hurt |
| use SCP ControlNet instead of vanilla ControlNet for expansion | dataset expansion across segmentation architectures · Crack500, DeepLabV3+ | mIoU: 76.8 -> 78.1 | Improved |
| use SCP ControlNet vs CE (GAN-based) with DeepLabV3+ on XCAD | dataset expansion · XCAD, DeepLabV3+ | mIoU: 85.7 (CE) vs 84.3 | Hurt |
| use SCP ControlNet vs CE for F1 on CHASEDB1 | dataset expansion, UNet · CHASEDB1 | F1: 81.5 (CE) vs 81.2 | No clear effect |
| expansion with synthetic data on Mask2Former for DRIVE and CHUAC | Mask2Former · DRIVE and CHUAC | mIoU: DRIVE 66.3 -> 67.1; CHUAC 50.1 -> 57.1 with SCP ControlNet | Mixed |
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.
- RandAugment — RandAugment: Practical Automated Data Augmentation with a Reduced Search Space (NeurIPS 2020) · 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
- From Reconstruction to Decision: A Post-Encoder Plug-in Adapter for Curvilinear Segmentation (ECCV 2026 · similarity 0.88)
- Context-Aware Spatio-Recurrent Curvilinear Structure Segmentation (CVPR 2019 · similarity 0.88)
- Leveraging Depth and Language for Open-Vocabulary Domain-Generalized Semantic Segmentation (NeurIPS 2025 · similarity 0.85)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.84)
- AdapToPASS: Ambiguity-aware Adaptive Spherical Transformer for Panoramic Semantic Segmentation (NeurIPS 2026 · similarity 0.84)
- econSG: Efficient and Multi-view Consistent Open-Vocabulary 3D Semantic Gaussians (ICLR 2025 · 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.
