MicroViTv2: Beyond the FLOPS for Edge Energy-Friendly Vision Transformers
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866909031321305088 |
|---|---|
| author | Setyawan, Novendra Sun, Chi-Chia Hsu, Mao-Hsiu Kuo, Wen-Kai Hsieh, Jun-Wei |
| author_facet | Setyawan, Novendra Sun, Chi-Chia Hsu, Mao-Hsiu Kuo, Wen-Kai Hsieh, Jun-Wei |
| contents | The Vision Transformer (ViT) achieves remarkable accuracy across visual tasks but remains computationally expensive for edge deployment. This paper presents MicroViTv2, a lightweight Vision Transformer optimized for real-device efficiency. Built upon the original MicroViT, the proposed model is designed based on reparameterized design, specifically Reparameterized Patch Embedding (RepEmbed) and Reparameterized Depth-Wise convolution mixer (RepDW) for faster inference, and introduces the Single Depth-Wise Transposed Attention (SDTA) to capture long-range dependencies with minimal redundancy. Despite slightly higher FLOPs, MicroViTv2 improves accuracy up to 0.5% compared to its predecessor and surpassing MobileViTv2, EdgeNeXt, and EfficientViT while maintaining fast inference and high energy efficiency on Jetson AGX Orin. Experiments on ImageNet-1K and COCO demonstrate that hardware-aware design and structural re-parameterization are key to achieving high accuracy and low energy consumption, validating the need to evaluate efficiency beyond FLOPs. Code is available at https://github.com/novendrastywn/MicroViT. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_10148 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | MicroViTv2: Beyond the FLOPS for Edge Energy-Friendly Vision Transformers Setyawan, Novendra Sun, Chi-Chia Hsu, Mao-Hsiu Kuo, Wen-Kai Hsieh, Jun-Wei Computer Vision and Pattern Recognition The Vision Transformer (ViT) achieves remarkable accuracy across visual tasks but remains computationally expensive for edge deployment. This paper presents MicroViTv2, a lightweight Vision Transformer optimized for real-device efficiency. Built upon the original MicroViT, the proposed model is designed based on reparameterized design, specifically Reparameterized Patch Embedding (RepEmbed) and Reparameterized Depth-Wise convolution mixer (RepDW) for faster inference, and introduces the Single Depth-Wise Transposed Attention (SDTA) to capture long-range dependencies with minimal redundancy. Despite slightly higher FLOPs, MicroViTv2 improves accuracy up to 0.5% compared to its predecessor and surpassing MobileViTv2, EdgeNeXt, and EfficientViT while maintaining fast inference and high energy efficiency on Jetson AGX Orin. Experiments on ImageNet-1K and COCO demonstrate that hardware-aware design and structural re-parameterization are key to achieving high accuracy and low energy consumption, validating the need to evaluate efficiency beyond FLOPs. Code is available at https://github.com/novendrastywn/MicroViT. |
| title | MicroViTv2: Beyond the FLOPS for Edge Energy-Friendly Vision Transformers |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2605.10148 |