Research map / Detection and segmentation / Open-set and few-shot detection
Balancing Logit Variation for Long-Tailed Semantic Segmentation
Yuchao Wang, Jingjing Fei, Haochen Wang, Wei Li, Tianpeng Bao, Liwei Wu, Rui Zhao, Yujun Shen
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 |
|---|---|---|---|
| Add category-balanced logit variation during training | Fully supervised segmentation models · Cityscapes val | mIoU: +0.72%, +0.35%, +0.55%, +0.47%, +0.24%, +0.43%, +1.20% | Improved |
| Add category-balanced logit variation during training | Self-Training · Cityscapes val, semi-supervised partitions 1/16, 1/8, 1/4, 1/2 | mIoU: +1.05%, +1.26%, +1.49%, +0.99% | Improved |
| Add category-balanced logit variation during training | DAFormer and HRDA · GTA5 → Cityscapes | mIoU: +3.1%, +1.3%, +1.1% | Improved |
| Add category-balanced logit variation during training | DAFormer and HRDA · SYNTHIA → Cityscapes | mIoU: +4.1%, +1.6%, +1.0% | Improved |
| Use Gaussian rather than no variation (ablation) | DAFormer · GTA5 → Cityscapes | mIoU: 55.9 -> 59.0 | Improved |
| Use Uniform logit variation (ablation) | DAFormer · GTA5 → Cityscapes | mIoU: 55.9 -> 58.2 | Improved |
| Use Beta logit variation (ablation) | DAFormer · GTA5 → Cityscapes | mIoU: 55.9 -> 57.9 | Improved |
| Use Exponential logit variation (ablation) | DAFormer · GTA5 → Cityscapes | mIoU: 55.9 -> 58.5 | Improved |
| Use σ=6 for logit variation (ablation) | DAFormer · GTA5 → Cityscapes | mIoU: 55.9 -> 59.0 | Improved |
| Use σ=4 for logit variation (ablation) | DAFormer · SYNTHIA → Cityscapes | mIoU: 52.7 -> 56.8 | Improved |
| Add temporal variation scheduling to BLV (ablation) | DAFormer + BLV · GTA5 → Cityscapes | mIoU: 69.6 -> 70.0 | Improved |
| Add random variation instead of a constant category adjustment (ablation) | BLV · GTA5 → Cityscapes | mIoU: 56.5 -> 59.0 | Improved |
Closest papers by meaning
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.88)
- Learning from Neighbors: Category Extrapolation for Long-Tail Learning (CVPR 2025 · similarity 0.87)
- Imagine by Reasoning: A Reasoning-Based Implicit Semantic Data Augmentation for Long-Tailed Classification (AAAI 2022 · similarity 0.87)
- Loss Max-Pooling for Semantic Image Segmentation (CVPR 2017 · similarity 0.87)
- Taking a Closer Look at Domain Shift: Category-Level Adversaries for Semantics Consistent Domain Adaptation (CVPR 2019 · similarity 0.86)
- Relaxed Class-consensus Consistency for Semi-supervised Semantic Segmentation (AAAI 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.
