Research map / Detection and segmentation / CNN features
Auto Seg-Loss: Searching Metric Surrogates for 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 |
|---|---|---|---|
| search a metric-specific surrogate for mIoU | DeepLabv3+ R101 · PASCAL VOC 2012 val | mIoU: 79.72 -> 80.97 | Helped |
| search a metric-specific surrogate for BIoU | DeepLabv3+ R101 · PASCAL VOC 2012 val | BIoU: 71.54 -> 79.27 | Helped |
| search a metric-specific surrogate for BF1 | DeepLabv3+ R101 · Cityscapes val | BF1: 72.20 -> 77.40 | Helped |
| add learnable parameterization to the continuous metric surrogate (ablation) | Auto Seg-Loss · PASCAL VOC 2012 | mIoU: 46.99 -> 80.64 | Helped |
| add the truth-table constraint (ablation) | Parameterized Auto Seg-Loss · PASCAL VOC 2012 | mIoU | Helped |
| add the monotonicity constraint (ablation) | Parameterized Auto Seg-Loss · PASCAL VOC 2012 | mIoU: 77.76 -> 80.64 | Helped |
| use 128 × 128 images instead of 256 × 256 for the search proxy (ablation) | Auto Seg-Loss search · PASCAL VOC 2012 | mIoU: 81.15 -> 80.97 | No clear effect |
| use the shorter 1000-iteration proxy schedule instead of 2000 iterations (ablation) | Auto Seg-Loss search · PASCAL VOC 2012 | mIoU: 80.56 -> 80.97 | No clear effect |
| use R50 rather than R101 for the lightweight search proxy (ablation) | Auto Seg-Loss search · PASCAL VOC 2012 | mIoU: 80.75 -> 80.97 | No clear effect |
| combine searched mIoU and BIoU surrogates | DeepLabv3+ R101 · PASCAL VOC 2012 val | BIoU: 79.27 -> 76.89 | Helped |
Closest papers by meaning
- AutoLoss-Zero: Searching Loss Functions From Scratch for Generic Tasks (CVPR 2022 · similarity 0.86)
- Severity-Aware Semantic Segmentation With Reinforced Wasserstein Training (CVPR 2020 · similarity 0.86)
- Relational Surrogate Loss Learning (ICLR 2022 · similarity 0.86)
- Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search (ICML 2026 · similarity 0.84)
- Complex Wavelet Mutual Information Loss: A Multi-Scale Loss Function for Semantic Segmentation (ICML 2025 · similarity 0.84)
- APSeg: Auto-Prompt Network for Cross-Domain Few-Shot Semantic Segmentation (CVPR 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.