Research map / Architectures and efficiency / CNN design
RepMLPNet: Hierarchical Vision MLP With Re-Parameterized Locality
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 |
|---|---|---|---|
| add 1×1 and 3×3 training-time convolution branches through Locality Injection (ablation) | ResMLP-12 · CIFAR-100 | Top-1 acc: 55.58 -> 65.09 | Helped |
| add 1×1 and 3×3 training-time convolution branches through Locality Injection (ablation) | RepMLPNet · CIFAR-100 | Top-1 acc: 59.07 -> 67.43 | Helped |
| add 1×1 and 3×3 Local Perceptron branches (ablation) | RepMLPNet-T224 · ImageNet | Top-1 acc: 74.33 -> 76.48 | Helped |
| add 1×1 and 3×3 Local Perceptron branches (ablation) | RepMLPNet-D256 · ImageNet | Top-1 acc: 78.58 -> 80.88 | Helped |
| add Global Perceptron (ablation) | RepMLP-T224, S=[1,4,16,128] · ImageNet | Top-1 acc: 75.78 -> 76.48 | Helped |
| increase share-sets from [1,4,16,128] to [2,8,32,256] (ablation) | RepMLP-T224 with Global Perceptron · ImageNet | Top-1 acc: 76.48 -> 77.19 | Helped |
| replace 3×3 downsampling layers with 5×5 layers (ablation) | RepMLPNet-D256 · Cityscapes val | mIoU: 76.27 -> 77.12 | Helped |
| use RepMLPNet-T256 instead of MLP-Mixer (our impl) at 256×256 input | ImageNet models · ImageNet | Top-1 acc: 77.0 -> 77.5 | Helped |
| remove DeiT-style training | ResMLP-S12 · ImageNet | Top-1 acc: 76.6 -> 67.7 | Hurt |
| use RepMLPNet-B256 instead of EfficientNet-B2 | ImageNet models · ImageNet | Top-1 acc: 77.4 -> 81.0 | Helped |
| replace RepMLPNet-D256 with ResNet-101 | Cityscapes segmentation backbone · Cityscapes val | mIoU: 76.27 -> 79.03 | Helped |
Closest papers by meaning
- Hire-MLP: Vision MLP via Hierarchical Rearrangement (CVPR 2022 · similarity 0.85)
- Locality Alignment Improves Vision-Language Models (ICLR 2025 · similarity 0.84)
- Make RepVGG Greater Again: A Quantization-Aware Approach (AAAI 2024 · similarity 0.84)
- CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective (CVPR 2024 · similarity 0.83)
- Data-independent Module-aware Pruning for Hierarchical Vision Transformers (ICLR 2024 · similarity 0.83)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · 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.