Research map / Label-efficient and robust learning / Noisy labels
Deep Neural Networks Learn Meta-Structures from Noisy Labels in Semantic Segmentation
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 |
|---|---|---|---|
| Train DeepLabv3 with random sampling or random flipping while retaining label meta-structure | DeepLabv3 · Cityscapes | mIoU(%): 64.8; 64.8, 64.5, 64.6; 64.7, 64.6, 64.7 | No clear effect |
| Train with randomly flipped labels retaining meta-structures instead of clean labels | U-Net and DeepLabv3+ · Binary segmentation datasets | DICE: maximum gap between final dice scores of CL and RCL is $7\%$ | No clear effect |
| Use randomly generated labels without ground-truth meta-structures instead of randomly flipped labels | U-Net · ER | DICE | Hurt |
| Perturb ground-truth meta-structures using dilation or erosion | U-Net · ER | DICE | Hurt |
| Replace full labels with one-pixel-wide skeleton labels, losing width information | U-Net · ER | DICE | Hurt |
| Add the extraction-of-meta-structure module to iGTT (ablation) | iGTT · ER | DICE(%): 73.96_{0.97} -> 78.84_{1.17} | Helped |
| Add the extraction-of-meta-structure module to iGTT (ablation) | iGTT · ER | AUC(%): 84.53_{2.52} -> 91.61_{1.04} | Helped |
| Add the extraction-of-meta-structure module to iGTT (ablation) | iGTT · ER | ACC(%): 81.16_{1.03} -> 85.41_{1.06} | Helped |
Closest papers by meaning
- On the Importance of Label Quality for Semantic Segmentation (CVPR 2018 · similarity 0.88)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.87)
- Latent Noise Segmentation: How Neural Noise Leads to the Emergence of Segmentation and Grouping (ICML 2024 · similarity 0.87)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.87)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.87)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.87)
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.