Research map / Detection and segmentation / CNN features
Loss Max-Pooling for Semantic Image Segmentation
Check 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 |
|---|---|---|---|
| Replace equal pixel-loss weighting with loss max-pooling | ResNet-101 segmentation model · Cityscapes validation | mean IoU: 73.63 ± 0.04 -> 75.06 ± 0.09 | Helped |
| Replace equal pixel-loss weighting with loss max-pooling | ResNet-101 segmentation model · Pascal VOC 2012 validation | mean IoU: 75.42 ± 0.04 -> 76.29 ± 0.02 | Helped |
| Add performance-dependent sampling to loss max-pooling training (ablation) | LMP · Cityscapes validation | mean IoU: 74.17 ± 0.03 -> 75.06 ± 0.09 | Helped |
| Add performance-dependent sampling to loss max-pooling training (ablation) | LMP · Pascal VOC 2012 validation | mean IoU: 76.01 ± 0.01 -> 76.29 ± 0.02 | Helped |
| Set p to 1.2 rather than 1.3 (ablation) | LMP · Cityscapes validation (efficient tiling) | mean IoU: 74.64 -> 74.60 | No clear effect |
| Set p to 1.0 rather than 1.3 (ablation) | LMP · Cityscapes validation (efficient tiling) | mean IoU: 74.64 -> 74.54 | No clear effect |
| Set p to 1.5 rather than 1.3 (ablation) | LMP · Cityscapes validation (efficient tiling) | mean IoU: 74.64 -> 74.02 | Hurt |
| Set m to select at least 10% rather than 25% of valid pixels (ablation) | LMP with p=1.3 · Cityscapes validation | mean IoU: 74.64 -> 74.09% ± 0.22 | Hurt |
| Set m to select at least 50% rather than 25% of valid pixels (ablation) | LMP with p=1.3 · Cityscapes validation | mean IoU: 74.64 -> 73.44 ± 0.21 | Hurt |
| Replace performance-based loss max-pooling with inverse median frequency weighting | ResNet-101 segmentation model · Cityscapes validation | mean IoU: 75.06 ± 0.09 -> 69.81 ± 0.08 | Hurt |
| Replace performance-based loss max-pooling with inverse median frequency weighting | ResNet-101 segmentation model · Pascal VOC 2012 validation | mean IoU: 76.29 ± 0.02 -> 74.93 ± 0.03 | Hurt |
| Use loss max-pooling rather than equal weighting | ResNet-101 segmentation model · Cityscapes validation, per-class IoU | classes improved: 18/19 classes | Helped |
Closest papers by meaning
- Building Detail-Sensitive Semantic Segmentation Networks With Polynomial Pooling (CVPR 2019 · similarity 0.91)
- Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation (CVPR 2021 · similarity 0.88)
- Weakly Supervised Semantic Segmentation Using Superpixel Pooling Network (AAAI 2017 · similarity 0.87)
- Detail-Preserving Pooling in Deep Networks (CVPR 2018 · similarity 0.87)
- A Unified Loss for Handling Inter-Class and Intra-Class Imbalance in Medical Image Segmentation (AAAI 2025 · similarity 0.86)
- Complex Wavelet Mutual Information Loss: A Multi-Scale Loss Function for Semantic Segmentation (ICML 2025 · similarity 0.85)
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.