Research map / Detection and segmentation / CNN features
AD-GBC: Anisotropic Granular-Ball Skip-Connection Refiner for UNet-Based Medical Image Segmentation
Xiya Shen, Qinglin Zhao, Li Feng
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 isotropic scalar scale with anisotropic vector scale in GBC (ablation) | Rolling-UNet-L + GBC · GlaS | IoU: 88.12 -> 88.28 | Improved |
| add isotropic GBC without regularizers (ablation) | Rolling-UNet-L · GlaS | IoU: 87.99 -> 88.12 | No clear effect |
| add Wasserstein diversity loss (ablation) | anisotropic GBC · GlaS | IoU: 88.28 -> 88.70 | Improved |
| add scale consistency loss (ablation) | anisotropic GBC · GlaS | IoU: 88.28 -> 88.78 | Improved |
| use both regularizers together (ablation) | anisotropic GBC · GlaS | IoU: 88.78 -> 89.63 | Improved |
| add AD-GBC to U-KAN | U-KAN · BUSI | IoU: 65.76 -> 69.86 | Improved |
| add AD-GBC to Rolling-UNet-L | Rolling-UNet-L · Average over four datasets | IoU: 81.43 -> 82.77 | Improved |
| add AD-GBC to U-KAN | U-KAN · CVC-ClinicDB and ISIC17 | IoU: 84.94 -> 85.03 (CVC); 84.52 -> 84.59 (ISIC17) | No clear effect |
| add AD-GBC to Rolling-UNet-S | Rolling-UNet-S · BUSI | IoU: 65.38 -> 65.62 | No clear effect |
| add AD-GBC module (cost) | Rolling-UNet-L · Efficiency | GFLOPs: 262.96 -> 304.32 | Hurt |
| raise temperature tau to 1.0 at K=32 (ablation) | AD-GBC · BUSI | IoU: 68.09 -> 70.71 (tau 0.01 -> 1.0) | Improved |
| raise tau to 1.0 at K=64 (ablation) | AD-GBC · BUSI | IoU: 69.65 -> 66.30 (tau 0.01 -> 1.0) | Hurt |
Closest papers by meaning
- UniPro: Unified Multi-Mode Medical Image Segmentation from 2D Images to 3D Volumes via Propagation (NeurIPS 2026 · similarity 0.85)
- FuseUNet: A Multi-Scale Feature Fusion Method for U-like Networks (ICML 2025 · similarity 0.85)
- Upping the Game: How 2D U-Net Skip Connections Flip 3D Segmentation (NeurIPS 2024 · similarity 0.85)
- Rethinking U-Net: Task-Adaptive Mixture of Skip Connections for Enhanced Medical Image Segmentation (AAAI 2025 · similarity 0.85)
- Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It (NeurIPS 2026 · similarity 0.84)
- MAPSeg: Unified Unsupervised Domain Adaptation for Heterogeneous Medical Image Segmentation Based on 3D Masked Autoencoding and Pseudo-Labeling (CVPR 2024 · 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.
