Research map / Detection and segmentation / CNN features
MutualNet: Adaptive ConvNet via Mutual Learning from Network Width and Resolution
Taojiannan Yang, Sijie Zhu, Chen Chen, Shen Yan, Mi Zhang, Andrew Willis
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| feed different input resolutions to sub-networks with width-resolution mutual learning, replacing constant 224 input | US-Net (MobileNet v1) · ImageNet, 150 MFLOPs | top-1 accuracy (%): 62.9 -> 65.6 | Improved |
| add mutual learning from width and resolution | US-Net (MobileNet v1) · ImageNet, 1.0x-224 | top-1 accuracy (%): 71.7 -> 72.4 | Improved |
| add mutual learning from width and resolution | US-Net (MobileNet v2) · ImageNet, 1.0x-224 | top-1 accuracy (%): 71.5 -> 72.9 | Improved |
| add multi-scale data augmentation to MobileNet v2 baseline (ablation) | MobileNet v2 · ImageNet, 1.0x-224 | top-1 accuracy (%): 71.8 -> 72.0 | No clear effect |
| add width-resolution mutual learning instead of multi-scale data augmentation (ablation) | MobileNet v2 · ImageNet, 1.0x-224 | top-1 accuracy (%): 72.0 -> 72.9 | Improved |
| replace EfficientNet grid-searched scaling with MutualNet scaling | MobileNet v1 · ImageNet, 2.3B FLOPs | top-1 accuracy (%): 75.6 -> 77.1 | Improved |
| add mutual learning (width lower bound 0.9x) as training scheme | WideResNet-28-10 · CIFAR-100 | top-1 accuracy (%): 81.2 -> 83.8 | Improved |
| add mutual learning as training scheme | WideResNet-28-10 · CIFAR-10 | top-1 accuracy (%): 96.1 -> 97.2 | Improved |
| add mutual learning as training scheme | ResNet-50 · ImageNet | top-1 accuracy (%): 76.5 -> 78.6 | Improved |
| add knowledge distillation from a teacher network | ResNet-50 · ImageNet | top-1 accuracy (%): 76.5 -> 76.5 | No clear effect |
| raise width lower bound from 0.7x to 0.9x (ablation) | MutualNet (MobileNet v2) · ImageNet | accuracy-FLOPs | Mixed |
| apply different resolutions only at inference (US-Net+) instead of training with them (ablation) | US-Net · ImageNet MobileNet v1 | resolution selection | Hurt |
Closest papers by meaning
- MutualNet: Adaptive ConvNet via Mutual Learning From Different Model Configurations (TPAMI 2021 · similarity 0.95)
- Revisiting ResNets: Improved Training and Scaling Strategies (NeurIPS 2021 · similarity 0.87)
- EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks (ICML 2019 · similarity 0.87)
- Resolution Adaptive Networks for Efficient Inference (CVPR 2020 · similarity 0.86)
- Mutual Learning for SAM Adaptation: A Dual Collaborative Network Framework for Source-Free Domain Transfer (ICML 2025 · similarity 0.86)
- Once-for-All: Train One Network and Specialize it for Efficient Deployment (ICLR 2020 · 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.
