Research map / Label-efficient and robust learning / Noisy labels
Learning to Segment from Noisy Annotations: A Spatial Correction Approach
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use Spatial Correction instead of the strongest baseline | U-Net · JSRT Lung, S_E | DSC (%): 82.08\pm0.06 -> 91.62\pm2.85 | Helped |
| Use Spatial Correction instead of the strongest baseline | U-Net · JSRT Clavicle, S_E | DSC (%): 79.37\pm0.18 -> 78.11\pm1.35 | Hurt |
| Use Spatial Correction instead of the strongest baseline | U-Net · ISIC 2017, S_E | DSC (%): 74.06\pm0.92 -> 80.64\pm0.63 | Helped |
| Use Spatial Correction instead of the strongest baseline | 3D-UNet · Brats 2020, S_E | DSC (%): 70.69\pm0.72 -> 77.78\pm0.85 | Helped |
| Use Spatial Correction instead of the strongest baseline | U-Net · JSRT Clavicle, S_S | DSC (%): 82.49\pm0.30 -> 75.78\pm2.02 | Hurt |
| Use Spatial Correction instead of the strongest baseline | U-Net · ISIC 2017, S_S | DSC (%): 54.76\pm1.15 -> 75.97\pm1.73 | Helped |
| Use Spatial Correction instead of the strongest baseline | U-Net · JSRT Heart, S_M | DSC (%): 89.09\pm0.84 -> 87.14\pm2.15 | Hurt |
| Use Spatial Correction instead of the strongest baseline | 3D-UNet · Brats 2020, S_M | DSC (%): 75.03\pm0.77 -> 72.89\pm0.98 | Hurt |
| Use Spatial Correction instead of the strongest baseline | U-Net · Cityscapes | DSC (%): 76.35\pm0.13 -> 82.30 \pm 0.10 | Helped |
| Use Spatial Correction instead of the strongest baseline | U-Net · LIDC-IDRI | DSC (%): 50.40\pm0.15 -> 53.72 \pm 1.70 | Helped |
| Reduce the clean validation set size from 24 to 18, 12, 6, and 1 samples (ablation) | Spatial Correction on JSRT S_M · JSRT S_M | DSC (%) | No clear effect |
| Increase Markov noise level by increasing T (ablation) | Spatial Correction on JSRT · JSRT, increasing-noise experiment | DSC (%) | No clear effect |
Closest papers by meaning
- Disentangling Human Error from Ground Truth in Segmentation of Medical Images (NeurIPS 2020 · similarity 0.92)
- Adaptive Early-Learning Correction for Segmentation From Noisy Annotations (CVPR 2022 · similarity 0.92)
- Towards Fairness under Label Bias in Image Segmentation: Impact, Measurement and Mitigation (NeurIPS 2026 · similarity 0.87)
- Latent Noise Segmentation: How Neural Noise Leads to the Emergence of Segmentation and Grouping (ICML 2024 · similarity 0.87)
- Annotation Ambiguity Aware Semi-Supervised Medical Image Segmentation (CVPR 2025 · similarity 0.86)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.86)
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.