Research map / Detection and segmentation / CNN features
Hire-MLP: Vision MLP via Hierarchical Rearrangement
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 token-mixing MLPs with hierarchical inner-region and cross-region rearrangement | Hire-MLP · ImageNet-1K val | Top-1 accuracy: 82.1 vs. 81.3 | Helped |
| Use Hire-MLP-Small rather than CycleMLP-B2 | ImageNet-1K val · ImageNet-1K val | Top-1 accuracy: 82.1 vs. 81.6 | Helped |
| Use Hire-MLP-Large rather than the strongest listed ImageNet baseline | ImageNet-1K val · ImageNet-1K val | Top-1 accuracy: 83.8 vs. 84.0 | Hurt |
| Use Hire-MLP-Small as the RetinaNet backbone rather than Swin-T | RetinaNet · COCO val2017, RetinaNet 1× | AP: 41.5 -> 41.7 | Helped |
| Use Hire-MLP-Base as the UperNet backbone rather than Swin-T | UperNet · ADE20K val | Single-scale mIoU: 44.5 -> 48.3 | Helped |
| Use Hire-MLP-Large as the UperNet backbone rather than Swin-B | UperNet · ADE20K val | Multi-scale mIoU: 49.7 -> 49.9 | Helped |
| Add cross-region restoration (ablation) | Hire-MLP-Small · ImageNet-1K val | Top-1 accuracy: 81.70 -> 82.07 | Helped |
| Add cross-region rearrangement and restoration (ablation) | Hire-MLP-Small · ImageNet-1K val | Top-1 accuracy: 81.18 -> 82.07 | Helped |
| Add inner-region rearrangement and restoration (ablation) | Hire-MLP-Small · ImageNet-1K val | Top-1 accuracy: 80.17 -> 82.07 | Helped |
| Add the extra channel-only FC branch (ablation) | Hire-MLP-Small · ImageNet-1K val | Top-1 accuracy: 81.32 -> 82.07 | Helped |
| Use shifted cross-region communication instead of ShuffleNet-style channel shuffle (ablation) | Hire-MLP-Small · ImageNet-1K val | Top-1 accuracy: 80.90 -> 82.07 | Helped |
| Use two rather than three or four intermediate FC layers (ablation) | Hire-MLP-Small · ImageNet-1K val | Top-1 accuracy: 81.81 -> 82.07; 81.85 -> 82.07 | Helped |
Closest papers by meaning
- Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · similarity 0.83)
- MetaSlot: Break Through the Fixed Number of Slots in Object-Centric Learning (NeurIPS 2025 · similarity 0.82)
- REOrdering Patches Improves Vision Models (NeurIPS 2025 · similarity 0.82)
- PLACE: Adaptive Layout-Semantic Fusion for Semantic Image Synthesis (CVPR 2024 · similarity 0.82)
- Maximizing the Position Embedding for Vision Transformers with Global Average Pooling (AAAI 2025 · similarity 0.81)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.81)
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.