Research map / Detection and segmentation / CNN features
AS-MLP: An Axial Shifted MLP Architecture for Vision
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 |
|---|---|---|---|
| Use axial shifts in both spatial directions with parallel feature fusion (ablation) | AS-MLP-T · ImageNet-1K | Top-1 accuracy (%): 81.34 vs. 79.81 | Helped |
| Increase shift size from (1, 1) to (5, 5) (ablation) | AS-MLP-T · ImageNet-1K | Top-1 accuracy (%): 74.17 -> 81.34 | Helped |
| Use zero rather than reflect padding (ablation) | AS-MLP-T with shift (3, 3) · ImageNet-1K | Top-1 accuracy (%): 81.14 -> 81.26 | Helped |
| Increase dilation rate from 1 to 2 (ablation) | AS-MLP-T with shift (3, 3) · ImageNet-1K | Top-1 accuracy (%): 81.26 -> 80.50 | Hurt |
| Increase shift size from (5, 5) to (7, 7) (ablation) | AS-MLP-T · ImageNet-1K | Top-1 accuracy (%): 81.34 -> 81.32 | No clear effect |
| Increase shift size from (5, 5) to (9, 9) (ablation) | AS-MLP-T · ImageNet-1K | Top-1 accuracy (%): 81.34 -> 81.16 | Hurt |
| Use parallel rather than serial shift connections (ablation) | AS-MLP-T · ImageNet-1K | Top-1 accuracy (%): 81.28 -> 81.34 | Helped |
| Use a two-axis (5, 5) shift rather than a one-directional shift (ablation) | AS-MLP-T · ImageNet-1K | Top-1 accuracy (%): 78.37 -> 81.34; 78.45 -> 81.34 | Helped |
| Use AS-MLP rather than Swin-T | ImageNet-1K, 224×224 · ImageNet-1K | Top-1 accuracy (%): 81.3 vs. 81.3 | No clear effect |
| Use AS-MLP-B rather than Swin-B | ImageNet-1K, 224×224 · ImageNet-1K | Top-1 accuracy (%): 83.3 vs. 83.3 | No clear effect |
| Use AS-MLP-T rather than Swin-T | Mask R-CNN on COCO · COCO val2017, 3× | APb: 46.0 vs. 46.0 | No clear effect |
| Use AS-MLP-B rather than Swin-B | Cascade Mask R-CNN on COCO · COCO val2017, 3× | APb: 51.9 -> 51.5 | Hurt |
Closest papers by meaning
- MLP-Mixer: An all-MLP Architecture for Vision (NeurIPS 2021 · similarity 0.89)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.86)
- Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · similarity 0.83)
- MLP Can Be A Good Transformer Learner (CVPR 2024 · similarity 0.83)
- Vision Transformer Neural Architecture Search for Out-of-Distribution Generalization: Benchmark and Insights (NeurIPS 2024 · similarity 0.83)
- Linear Differential Vision Transformer: Learning Visual Contrasts via Pairwise Differentials (NeurIPS 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.