Research map / Detection and segmentation / CNN features
MobileOne: An Improved One Millisecond Mobile Backbone
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 |
|---|---|---|---|
| Use ReLU instead of Dynamic Shift-Max activation (ablation) | 30-layer CNN on iPhone12 · 30-layer CNN | Latency (ms): 57.04 -> 1.53 | Helped |
| Use ReLU instead of DynamicReLU-A activation (ablation) | 30-layer CNN on iPhone12 · 30-layer CNN | Latency (ms): 273.49 -> 1.53 | Helped |
| Add Squeeze-Excite blocks (ablation) | 30-layer CNN · iPhone12 | Latency (ms): 1.53 -> 2.10 | Hurt |
| Add skip connections (ablation) | 30-layer CNN · iPhone12 | Latency (ms): 1.53 -> 2.62 | Hurt |
| Add train-time re-parameterizable branches (ablation) | MobileOne-S0 · ImageNet-1K val | Top-1 accuracy: 69.6 -> 71.4 | Helped |
| Add train-time re-parameterizable branches (ablation) | MobileOne-S1 · ImageNet-1K val | Top-1 accuracy: 74.6 -> 75.9 | Helped |
| Vary trivial over-parameterization factor k (ablation) | MobileOne-S0 · ImageNet-1K val | Top-1 accuracy: 70.9, 70.7, 71.3, 71.4, 71.1 | Mixed |
| Vary trivial over-parameterization factor k (ablation) | MobileOne-S1 · ImageNet-1K val | Top-1 accuracy: 75.9, 75.7, 75.6, 75.6, 75.2 | Hurt |
| Add progressive learning (ablation) | MobileOne-S2 · ImageNet-1K val | Top-1 accuracy: 76.4 -> 76.8 | Helped |
| Anneal weight decay during training (ablation) | MobileOne-S2 · ImageNet-1K val | Top-1 accuracy: 76.8 -> 77.3 | Helped |
| Add EMA after progressive learning and annealed weight decay (ablation) | MobileOne-S2 · ImageNet-1K val | Top-1 accuracy: 77.3 -> 77.4 | Helped |
| Apply MobileOne training recipe to EfficientNet-B0 (ablation) | EfficientNet-B0 · ImageNet-1K val | Top-1 accuracy: 77.4 -> 77.8 | Helped |
Closest papers by meaning
- MnasNet: Platform-Aware Neural Architecture Search for Mobile (CVPR 2019 · similarity 0.90)
- ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices (CVPR 2018 · similarity 0.87)
- IFORMER: INTEGRATING CONVNET AND TRANSFORMER FOR MOBILE APPLICATION (ICLR 2025 · similarity 0.84)
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 2024 · similarity 0.84)
- FreeNet: Liberating Depth-Wise Separable Operations for Building Faster Mobile Vision Architectures (AAAI 2025 · similarity 0.84)
- SnapGen: Taming High-Resolution Text-to-Image Models for Mobile Devices with Efficient Architectures and Training (CVPR 2025 · 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.