Research map / Architectures and efficiency / Neural architecture search
HR-NAS: Searching Efficient High-Resolution Neural Architectures With Lightweight Transformers
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 the single branch with multi-branch architecture (ablation) | Cityscapes segmentation search space · Cityscapes validation | mIoU: 66.23 -> 68.65 | Helped |
| add MixConv with 3×3, 5×5, and 7×7 convolutions (ablation) | multi-branch search space · Cityscapes validation | mIoU: 68.65 -> 71.99 | Helped |
| add the lightweight Transformer (ablation) | multi-branch + MixConv search space · Cityscapes validation | mIoU: 71.99 -> 74.55 | Helped |
| use the proposed 2D positional map instead of sinusoidal encoding (ablation) | lightweight Transformer · Cityscapes validation | mIoU: 71.91 -> 74.22 | Helped |
| use the proposed 2D positional map instead of learned position embedding (ablation) | lightweight Transformer · Cityscapes validation | mIoU: 72.71 -> 74.22 | Helped |
| add 2D positional encoding to the decoder as well as the encoder (ablation) | lightweight Transformer · Cityscapes validation | mIoU: 74.22 -> 73.70 | Hurt |
| add the lightweight Transformer to inverted residual blocks | ShuffleNetV2 · ImageNet | Top-1 accuracy: 69.5 -> 70.1 | Helped |
| add the lightweight Transformer to inverted residual blocks | ShuffleNetV2 · Cityscapes | mIoU: 66.02 -> 67.31 | Helped |
| add the lightweight Transformer to inverted residual blocks | MobileNetV2 · ImageNet | Top-1 accuracy: 72.0 -> 72.8 | Helped |
| add the lightweight Transformer to inverted residual blocks | MobileNetV2 · Cityscapes | mIoU: 68.98 -> 70.17 | Helped |
| increase projection size from 8×8 to 16×16 (ablation) | lightweight Transformer · Cityscapes validation | mIoU: 74.22 -> 74.18 | No clear effect |
| use channel-wise rather than spatial tokens (ablation) | encoder + decoder Transformer · Cityscapes validation | mIoU: 73.54 -> 74.22 | Helped |
Closest papers by meaning
- Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation (CVPR 2019 · similarity 0.88)
- HRFormer: High-Resolution Vision Transformer for Dense Predict (NeurIPS 2021 · similarity 0.87)
- Vision-Oriented Lightweight Neural Architecture Search with Budget-Adaptive Evaluation (CVPR 2026 · similarity 0.87)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.87)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.86)
- No Pains, More Gains: Recycling Sub-Salient Patches for Efficient High-Resolution Image Recognition (CVPR 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.