Research map / Detection and segmentation / CNN features
Channelized Axial Attention – considering Channel Relation within Spatial Attention 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 |
|---|---|---|---|
| Integrate channel attention inside axial attention (ablation) | Axial Attention · PASCAL Context | mIOU (%): 50.27($\pm$0.2) -> 51.06($\pm$0.2) | Helped |
| Replace sequential standard SE with spatially varying channelization (ablation) | Axial Attention · PASCAL Context | mIOU (%): 50.37($\pm$0.2) -> 51.06($\pm$0.2) | Helped |
| Increase channel-attention depth from 1 to 5 layers (ablation) | CAA · PASCAL Context | mIOU (%): 50.75($\pm$0.2) -> 51.06($\pm$0.2) | Helped |
| Increase channel-attention depth from 5 to 7 layers (ablation) | CAA · PASCAL Context | mIOU (%): 51.06($\pm$0.2) -> 50.40($\pm$0.3) | Hurt |
| Reduce hidden channel width from 128 to 64 (ablation) | CAA with 5 layers · PASCAL Context | mIOU (%): 51.06($\pm$0.2) -> 50.12($\pm$0.2) | Hurt |
| Increase hidden channel width from 128 to 256 (ablation) | CAA with 5 layers · PASCAL Context | mIOU (%): 51.06($\pm$0.2) -> 50.35($\pm$0.4) | Hurt |
| Apply channelization to self-attention (ablation) | Self Attention · PASCAL Context, OS 16 | mIOU (%): 50.42 -> 51.09 | Helped |
| Use multi-scale and left-right flip at inference (ablation) | CAA · PASCAL Context, OS 8 | mIOU (%): 52.73($\pm$0.1) -> 54.05($\pm$0.1) | Helped |
| Add auxiliary loss during training (ablation) | CAA · PASCAL Context, OS 16 | mIOU (%): 51.06($\pm$0.2) -> 51.80($\pm$0.2) | Helped |
| Use OS 8 instead of OS 16 (ablation) | CAA · PASCAL Context | mIOU (%): 51.06($\pm$0.2) -> 52.73($\pm$0.1) | Helped |
| Use CAA rather than the strongest listed baseline | ResNet-101 semantic segmentation · PASCAL Context test | mIOU (%): 54.0 -> 55.0 | Helped |
| Use CAA rather than the strongest listed baseline | ResNet-101 semantic segmentation · COCO-Stuff 10K test | mIOU (%): 39.9 -> 41.2 | Helped |
Closest papers by meaning
- Fully Attentional Network for Semantic Segmentation (AAAI 2022 · similarity 0.88)
- Dual Attention Network for Scene Segmentation (CVPR 2019 · similarity 0.86)
- Attention (as Discrete-Time Markov) Chains (NeurIPS 2025 · similarity 0.85)
- Seg4Diff: Unveiling Open-Vocabulary Semantic Segmentation in Text-to-Image Diffusion Transformers (NeurIPS 2025 · similarity 0.83)
- Elliptical Attention (NeurIPS 2024 · similarity 0.83)
- Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas (ICML 2025 · 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.