Research map / Architectures and efficiency / CNN design
MLP-Mixer: An all-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 |
|---|---|---|---|
| Pre-train on larger datasets | MLP-Mixer · ImageNet transfer | top-1 accuracy: 76.44 -> 86.32 | Helped |
| Add modern augmentation and regularization during ImageNet pre-training | MLP-Mixer · ImageNet | top-1 accuracy | Helped |
| Use more pre-training data | Mixer-L/16 · ImageNet linear 5-shot transfer | top-1 accuracy | Helped |
| Untie token-mixing MLP parameters across channels (ablation) | MLP-Mixer token-mixing layers · JFT-300M pre-training | accuracy | No clear effect |
| Group neighboring channels by concatenation for token mixing (ablation) | MLP-Mixer · ImageNet linear 5-shot | top-1 accuracy: less than 1--2% | Helped |
| Group multiple learned channel views for token mixing (ablation) | Mixer-S/32 · ImageNet | top-5 accuracy: 3--4% | Helped |
| Group multiple learned channel views for token mixing (ablation) | larger MLP-Mixer scales · ImageNet | top-5 accuracy | No clear effect |
| Use a pyramid that reduces token count in later layers (ablation) | MLP-Mixer · ImageNet and ImageNet-21k transfer | accuracy | No clear effect |
| Apply mixup during fine-tuning (ablation) | JFT-300M-pretrained MLP-Mixer · downstream classification | accuracy | No clear effect |
| Apply Polyak averaging during fine-tuning (ablation) | JFT-300M-pretrained MLP-Mixer · downstream classification | accuracy | No clear effect |
| Use inception cropping during fine-tuning (ablation) | JFT-300M-pretrained MLP-Mixer · downstream classification | accuracy | No clear effect |
| Globally permute image pixels | MLP-Mixer versus ResNet50x1 · ImageNet linear 5-shot | top-1 accuracy drop: ~45% drop; ~75% drop | Mixed |
Closest papers by meaning
- MAXIM: Multi-Axis MLP for Image Processing (CVPR 2022 · similarity 0.90)
- Understanding MLP-Mixer as a wide and sparse MLP (ICML 2024 · similarity 0.88)
- LSNet: See Large, Focus Small (CVPR 2025 · similarity 0.86)
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · similarity 0.85)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.85)
- FreeNet: Liberating Depth-Wise Separable Operations for Building Faster Mobile Vision Architectures (AAAI 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.