Research map / Detection and segmentation / Weakly supervised detection and segmentation
Harnessing SSL for Segmentation in 3D Microscopy with Noisy Labels and Hard Patches
Lingtong Xu, Ahmadreza Attarpour, Shruti Patel, Fengqing Yu, Matthew Rozak, Bojana Stefanovic, Anne Martel, Maged Goubran
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 |
|---|---|---|---|
| initialize with frozen µ3DINO instead of random ViT (ablation) | UNETR segmentation pipeline · TPEMN (100% data) | Dice (%): 73.2 -> 77.1 | Improved |
| initialize with frozen µ3DINO instead of random ViT (ablation) | UNETR segmentation pipeline · TPEMV (100% data) | Dice (%): 77.4 -> 81.2 | Improved |
| initialize with frozen µ3DINO instead of random ViT (ablation) | UNETR segmentation pipeline · TPEMV (10% data) | Dice (%): 64.6 -> 70.0 | Improved |
| initialize with frozen µ3DINO instead of random ViT (ablation) | UNETR segmentation pipeline · LSFMA (2% data) | Dice (%): 65.1 -> 69.9 | Improved |
| initialize with µ3DINO instead of random ViT (ablation) | UNETR segmentation pipeline · four real-world datasets, average | Dice (%): 5.8% and 3.9% Dice improvement | Improved |
| add µDivSeg noisy-label and hard-patch guidance on top of µ3DINO (ablation) | µ3DINO · TPEMN (100% data) | Dice (%): 77.1 -> 79.5 | Improved |
| add µDivSeg guidance on top of µ3DINO (ablation) | µ3DINO · LSFMA (100% data) | Dice (%): 73.3 -> 76.8 | Improved |
| add µDivSeg guidance on top of µ3DINO (ablation) | µ3DINO · LSFMN (100% data) | Dice (%): 76.1 -> 75.2 | Hurt |
| add µDivSeg guidance on top of µ3DINO (ablation) | µ3DINO · LSFMA (2% data) | Dice (%): 69.9 -> 69.4 | No clear effect |
| use µDivSeg instead of standard pipeline with µ3DINO (ablation) | µ3DINO · real-world datasets, average | Dice (%): 1.1% and 1.9% increase | Improved |
| use µDivSeg instead of GuidedNetµ | GuidedNetµ · real-world datasets (100% data), average | Dice (%): +6.4% | Improved |
| use µDivSeg instead of L2Bµ | L2Bµ · real-world datasets (100% data), average | Dice (%): +9.4% | Improved |
Closest papers by meaning
- Spatial-SAM: Spatially Consistent 3D Electron Microscopy Segmentation with SDF Memory and Semi-Supervised Learning (CVPR 2026 · similarity 0.86)
- V2V3D: View-to-View Denoised 3D Reconstruction for Light Field Microscopy (CVPR 2025 · similarity 0.83)
- A Unified Degradation-Robust Approach to SSL and UDA for 3D Medical Images (AAAI 2025 · similarity 0.83)
- $\epsilon$-Seg: Sparsely Supervised Semantic Segmentation of Microscopy Data (NeurIPS 2025 · similarity 0.83)
- VISTA3D: A Unified Segmentation Foundation Model For 3D Medical Imaging (CVPR 2025 · similarity 0.83)
- SR-CACO-2: A Dataset for Confocal Fluorescence Microscopy Image Super-Resolution (NeurIPS 2024 · similarity 0.83)
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.
