Research map / Detection and segmentation / CNN features
An Image Patch Is a Wave: Phase-Aware Vision MLP
Yehui Tang, Kai Han, Jianyuan Guo, Chang Xu, Yanxi Li, Chao Xu, Yunhe Wang
AI-extracted. The measured effects below were read by AI from the paper's full text (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 fixed real-valued token mixing with input-dependent dynamic phase modulation (ablation) | Wave-MLP-T* · ImageNet | Top-1 accuracy: 78.8 -> 80.1 | Improved |
| Add input-dependent dynamic phase rather than static learned phase (ablation) | Wave-MLP-T* · ImageNet | Top-1 accuracy: 79.3 -> 80.1 | Improved |
| Use depth-wise convolution rather than identity projection to estimate phase (ablation) | Wave-MLP-T* · ImageNet | Top-1 accuracy: 79.3 -> 80.1 | Improved |
| Use channel-FC rather than depth-wise convolution to estimate phase (ablation) | Wave-MLP-T* · ImageNet | Top-1 accuracy: 80.1 -> 80.6 | Improved |
| Increase local token-mixing window size from 3 to 7 (ablation) | Wave-MLP-T* · ImageNet | Top-1 accuracy: 79.7 -> 80.1 | Improved |
| Use all-token mixing instead of window size 7 (ablation) | Wave-MLP-T* · ImageNet | Top-1 accuracy: 80.1 -> 80.0 | No clear effect |
| Use Wave-MLP-S instead of Swin-T as the ImageNet classifier | ImageNet classification · ImageNet val | Top-1 accuracy: 81.3 -> 82.6 | Improved |
| Use Wave-MLP-S instead of Swin-T as the RetinaNet backbone | RetinaNet 1× · COCO val2017 | AP: 41.5 -> 43.4 | Improved |
| Use Wave-MLP-S instead of Swin-T as the Mask R-CNN backbone | Mask R-CNN 1× · COCO val2017 | Box AP: 42.2 -> 44.0 | Improved |
| Use Wave-MLP-S instead of Swin-T as the Mask R-CNN backbone | Mask R-CNN 1× · COCO val2017 | Mask AP: 39.1 -> 40.0 | Improved |
| Use Wave-MLP-M instead of CycleMLP-B3 as the Semantic FPN backbone | Semantic FPN · ADE20K val | mIoU: 44.5 -> 46.8 | Improved |
Closest papers by meaning
- MLP-Mixer: An all-MLP Architecture for Vision (NeurIPS 2021 · similarity 0.85)
- WaveFormer: Frequency-Time Decoupled Vision Modeling with Wave Equation (AAAI 2026 · similarity 0.85)
- AS-MLP: An Axial Shifted MLP Architecture for Vision (ICLR 2022 · similarity 0.84)
- Scaling Laws in Patchification: An Image Is Worth 50,176 Tokens And More (ICML 2025 · similarity 0.84)
- ImagePiece: Content-aware Re-tokenization for Efficient Image Recognition (AAAI 2025 · similarity 0.83)
- Patch-as-Decodable-Token: Towards Unified Multi-Modal Vision Tasks in MLLMs (ICLR 2026 · similarity 0.82)
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.
