Research map / Detection and segmentation / CNN features
Multi-Scale High-Resolution Vision Transformer 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 |
|---|---|---|---|
| Use a multi-branch high-resolution architecture with cross-resolution fusion | ViT backbone · Cityscapes val | mIoU: +1.18 | Helped |
| Add key-value projection sharing (ablation) | HRViT-b1 · Cityscapes val | mIoU: 81.00 -> 81.63 | Helped |
| Use efficient patch embedding instead of the CONV-based counterpart (ablation) | HRViT-b1 · Cityscapes val | mIoU: 81.18 -> 81.63 | No clear effect |
| Add MixCFN instead of original FFNs (ablation) | HRViT-b1 · Cityscapes val | mIoU: 80.52 -> 81.63 | Helped |
| Add the parallel convolution path (ablation) | HRViT-b1 · Cityscapes val | mIoU: 80.82 -> 81.63 | Helped |
| Add the extra Hardswish nonlinearity and BatchNorm (ablation) | HRViT-b1 · Cityscapes val | mIoU: 81.12 -> 81.63 | Helped |
| Use dense fusion rather than sparse fusion (ablation) | HRViT-b1 · Cityscapes val | mIoU: 81.26 -> 81.63 | Helped |
| Add the diversity-enhanced shortcut (ablation) | HRViT-b1 · Cityscapes val | mIoU: 81.38 -> 81.63 | Helped |
| Add all block optimization techniques (ablation) | HRViT-b1 · Cityscapes val | mIoU: 80.45 -> 81.63 | Helped |
| Increase MR and LR attention window size from 7 to 9 (ablation) | HRViT-b3 · Cityscapes val | mIoU: +0.34 mIoU | Helped |
| Increase MR and LR attention window size from 9 to 15 (ablation) | HRViT-b3 · Cityscapes val | mIoU: 0.26 mIoU drop | Hurt |
| Use a more concentrated third-branch block assignment, from 6-6-6-2 to 17-1-1-1 (ablation) | HRViT-b1 · Cityscapes val | mIoU: 81.63 -> 81.27 | Hurt |
Closest papers by meaning
- MPViT: Multi-Path Vision Transformer for Dense Prediction (CVPR 2022 · similarity 0.89)
- AiluRus: A Scalable ViT Framework for Dense Prediction (NeurIPS 2023 · similarity 0.89)
- SegViT: Semantic Segmentation with Plain Vision Transformers (NeurIPS 2022 · similarity 0.88)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.88)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation (AAAI 2024 · similarity 0.87)
- Your ViT is Secretly an Image Segmentation Model (CVPR 2025 · similarity 0.86)
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.