Research map / Architectures and efficiency / CNN design
MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer
Sachin Mehta, Mohammad Rastegari
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 |
|---|---|---|---|
| Use MobileViT-S instead of MobileNetv2 as the DeepLabv3 backbone | DeepLabv3 · PASCAL VOC 2012 validation | mIOU: 75.7 -> 77.1 | Improved |
| Use MobileViT-S instead of MNASNet as the SSDLite backbone | SSDLite · MS-COCO validation | mAP: 23.0 -> 27.7 | Improved |
| Use multi-scale sampler instead of standard sampler | MobileNetv2-1.0 · ImageNet-1k | Top-1 accuracy: 72.1 -> 73.3 | Improved |
| Use multi-scale sampler instead of standard sampler | ResNet-50 · ImageNet-1k | Top-1 accuracy: 77.1 -> 78.6 | Improved |
| Add the MobileViT-block skip connection (ablation) | MobileViT-S · ImageNet-1k | Top-1 accuracy: +0.5% | Improved |
| Set patch sizes to 3,3,3 instead of 2,2,2 (ablation) | MobileViT-S · ImageNet-1k | Top-1 accuracy: 78.4 -> 78.5 | Improved |
| Set patch sizes to 4,4,4 instead of 2,2,2 (ablation) | MobileViT-S · ImageNet-1k | Top-1 accuracy: 78.4 -> 77.6 | Hurt |
| Set patch sizes to 8,4,2 instead of 2,2,2 (ablation) | MobileViT-S · ImageNet-1k | Top-1 accuracy: 78.4 -> 77.3 | Hurt |
| Add label smoothing (ablation) | MobileViT-S · ImageNet-1k | Top-1 accuracy: 78.0 -> 78.3 | Improved |
| Add EMA after label smoothing (ablation) | MobileViT-S · ImageNet-1k | Top-1 accuracy: 78.3 -> 78.4 | No clear effect |
| Use weight decay values from 0.1 to 0.0001 (ablation) | MobileViT-S · ImageNet-1k | validation accuracy | No clear effect |
Closest papers by meaning
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 2024 · similarity 0.93)
- SHViT: Single-Head Vision Transformer with Memory Efficient Macro Design (CVPR 2024 · similarity 0.89)
- GradQ-ViT: Robust and Efficient Gradient Quantization for Vision Transformers (AAAI 2025 · similarity 0.89)
- FasterViT: Fast Vision Transformers with Hierarchical Attention (ICLR 2024 · similarity 0.88)
- IFORMER: INTEGRATING CONVNET AND TRANSFORMER FOR MOBILE APPLICATION (ICLR 2025 · similarity 0.88)
- Real-time Core-Periphery Guided ViT with Smart Data Layout Selection on Mobile Devices (NeurIPS 2024 · 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.
