Research map / Architectures and efficiency / CNN design
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
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 |
|---|---|---|---|
| Balance depth, width, and resolution with compound scaling | MobileNetV1 · ImageNet validation | top-1 accuracy: 70.6% -> 75.6% | Helped |
| Balance depth, width, and resolution with compound scaling | MobileNetV2 · ImageNet validation | top-1 accuracy: 72.0% -> 77.4% | Helped |
| Balance depth, width, and resolution with compound scaling | ResNet-50 · ImageNet validation | top-1 accuracy: 76.0% -> 78.8% | Helped |
| Use compound scaling instead of single-dimension scaling (ablation) | EfficientNet-B0 · ImageNet validation | top-1 accuracy: 81.1% vs. 79.0%, 78.9%, 79.1% | Helped |
| Scale depth alone rather than using compound scaling (ablation) | EfficientNet-B0 · ImageNet validation | top-1 accuracy: 81.1% -> 79.0% | Hurt |
| Scale width alone rather than using compound scaling (ablation) | EfficientNet-B0 · ImageNet validation | top-1 accuracy: 81.1% -> 78.9% | Hurt |
| Scale resolution alone rather than using compound scaling (ablation) | EfficientNet-B0 · ImageNet validation | top-1 accuracy: 81.1% -> 79.1% | Hurt |
| Use compound scaling rather than width-only scaling (ablation) | MobileNetV1 · ImageNet validation | top-1 accuracy: 74.2% -> 75.6% | Helped |
| Use compound scaling rather than resolution-only scaling (ablation) | MobileNetV1 · ImageNet validation | top-1 accuracy: 72.7% -> 75.6% | Helped |
| Use compound scaling rather than depth-only scaling (ablation) | MobileNetV2 · ImageNet validation | top-1 accuracy: 76.8% -> 77.4% | Helped |
| Use compound scaling rather than width-only scaling (ablation) | MobileNetV2 · ImageNet validation | top-1 accuracy: 76.4% -> 77.4% | Helped |
| Use compound scaling rather than resolution-only scaling (ablation) | MobileNetV2 · ImageNet validation | top-1 accuracy: 74.8% -> 77.4% | Helped |
Closest papers by meaning
- Fast and Accurate Model Scaling (CVPR 2021 · similarity 0.92)
- EfficientNetV2: Smaller Models and Faster Training (ICML 2021 · similarity 0.91)
- Revisiting ResNets: Improved Training and Scaling Strategies (NeurIPS 2021 · similarity 0.90)
- ScaleNet - Improve CNNs through Recursively Rescaling Objects (AAAI 2020 · similarity 0.88)
- Rethinking the Inception Architecture for Computer Vision (CVPR 2016 · similarity 0.88)
- Refining Architectures of Deep Convolutional Neural Networks (CVPR 2016 · similarity 0.88)
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.