Research map / Detection and segmentation / CNN features
Partial Class Activation Attention for Semantic 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 image-level CAM pooling with region-level partial CAM using S=4 (ablation) | PCAA · Cityscapes validation | mIoU: 77.47 -> 79.22 | Helped |
| Increase pooling size from S=4 to S=8 (ablation) | PCAA · Cityscapes validation | mIoU: 79.22 -> 79.00 | Hurt |
| Increase pooling size from S=4 to S=16 (ablation) | PCAA · Cityscapes validation | mIoU: 79.22 -> 78.25 | Hurt |
| Use cascaded pooling sizes S=4 and S=8 (ablation) | PCAA · Cityscapes validation | mIoU: 79.22 -> 79.29 | Helped |
| Use local class centers for both keys and values instead of local keys and global values (ablation) | PCAA · Cityscapes validation | mIoU: 79.22 -> 78.89 | Hurt |
| Use global class centers only instead of local keys and global values (ablation) | PCAA · Cityscapes validation | mIoU: 79.22 -> 78.68 | Hurt |
| Add a graph convolution unit to local center generation (ablation) | PCAA using local centers · Cityscapes validation | mIoU: 75.36 -> 78.89 | Helped |
| Add PCAA to the FCN baseline | Cityscapes validation model with ResNet-50 · Cityscapes validation | mIoU: 74.68 -> 79.22 | Helped |
| Use PCAA instead of OCR attention | Cityscapes validation model with ResNet-50 · Cityscapes validation | mIoU: 78.86 -> 79.22 | Helped |
| Use PCAA instead of DNL | ResNet-101 segmentation model · Cityscapes test | mIoU: 82.0 -> 82.3 | Helped |
| Use PCAA instead of DNL | ResNet-101 segmentation model · Pascal Context test | mIoU: 54.8 -> 55.6 | Helped |
| Use PCAA instead of STLNet | Segmentation model on ADE20K · ADE20K validation | mIoU: 46.48 -> 46.74 | Helped |
Closest papers by meaning
- Embedded Discriminative Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2021 · similarity 0.89)
- MoRe: Class Patch Attention Needs Regularization for Weakly Supervised Semantic Segmentation (AAAI 2025 · similarity 0.89)
- L2G: A Simple Local-to-Global Knowledge Transfer Framework for Weakly Supervised Semantic Segmentation (CVPR 2022 · similarity 0.89)
- Extracting Class Activation Maps From Non-Discriminative Features As Well (CVPR 2023 · similarity 0.88)
- Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation (CVPR 2025 · similarity 0.87)
- SFC: Shared Feature Calibration in Weakly Supervised Semantic Segmentation (AAAI 2024 · 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.