Research map / Detection and segmentation / CNN features
CAR: Class-Aware Regularizations for Semantic Segmentation
Ye Huang, Di Kang, Liang Chen, Xuefei Zhe, Wenjing Jia, Linchao Bao, Xiangjian He
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 intra-class center-to-pixel loss (ablation) | ResNet-50 + Self-Attention · Pascal Context | mIoU: 48.32 -> 49.62 | Improved |
| add inter-class center-to-center loss (ablation) | ResNet-50 + Self-Attention + intra-c2p · Pascal Context | mIoU: 49.17 -> 49.79 | Improved |
| add inter-class center-to-pixel loss (ablation) | ResNet-50 + Self-Attention + intra-c2p + inter-c2c · Pascal Context | mIoU: 49.79 -> 50.01 | Improved |
| add all CAR losses | ResNet-50 + Self-Attention · Pascal Context | mIoU: 48.32 -> 50.50 | Improved |
| add intra-class center-to-pixel loss (ablation) | Swin-Tiny + UperNet · Pascal Context | mIoU: 49.62 -> 49.01 | Hurt |
| add inter-class center-to-center loss (ablation) | Swin-Tiny + UperNet + intra-c2p · Pascal Context | mIoU: 49.01 -> 50.63 | Improved |
| add inter-class center-to-pixel loss (ablation) | Swin-Tiny + UperNet + intra-c2p + inter-c2c · Pascal Context | mIoU: 50.63 -> 50.26 | Hurt |
| add CAR | Self-Attention ResNet-50 · COCO-Stuff10K | mIoU: 34.35 -> 36.58 | Improved |
| add CAR | DeepLabV3 ResNet-50 · COCO-Stuff10K | mIoU: 34.96 -> 35.13 | No clear effect |
| use batch class center instead of image class center (ablation) | CAR on ResNet-50 + Self-Attention · Pascal Context | mIoU: 49.78 -> 50.50 | Improved |
| use batch class center instead of image class center (ablation) | CAR on Swin-Tiny + UperNet · Pascal Context | mIoU: 49.45 -> 50.78 | Improved |
| add moving average of class center, decay 0.8 (ablation) | CAR on ResNet-50 + Self-Attention · Pascal Context | mIoU: 50.50 -> 49.80 | Hurt |
Closest papers by meaning
- Boosting Semantic Segmentation from the Perspective of Explicit Class Embeddings (ICCV 2023 · similarity 0.89)
- Incrementer: Transformer for Class-Incremental Semantic Segmentation With Knowledge Distillation Focusing on Old Class (CVPR 2023 · similarity 0.88)
- ClassFormer: Exploring Class-Aware Dependency with Transformer for Medical Image Segmentation (AAAI 2023 · similarity 0.88)
- Cs2K: Class-specific and Class-shared Knowledge Guidance for Incremental Semantic Segmentation (ECCV 2024 · similarity 0.88)
- Unbiased Subclass Regularization for Semi-Supervised Semantic Segmentation (CVPR 2022 · similarity 0.88)
- CSL: Class-Agnostic Structure-Constrained Learning for Segmentation Including the Unseen (AAAI 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.
