Research map / Architectures and efficiency / Attention and transformers
A Survey on Efficient Vision Transformers: Algorithms, Techniques, and Performance Benchmarking
Lorenzo Papa, Paolo Russo, Irene Amerini, Luping Zhou
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 |
|---|---|---|---|
| apply Castling linear attention training | MViTv2-T · ImageNet-1k classification (CA models) | Top-1 (%): 82.3 -> 84.1 | Improved |
| apply Castling linear attention training | MViTv2-S · ImageNet-1k classification (CA models) | Top-1 (%): 83.6 -> 84.6 | Improved |
| apply Castling linear attention training | MViTv2-B · ImageNet-1k classification (CA models) | Top-1 (%): 84.4 -> 85.0 | Improved |
| scale from Ortho-T to Ortho-S | Ortho · ImageNet-1k classification (CA models) | Top-1 (%): 74.0 -> 83.4 | Improved |
| scale from Ortho-S to Ortho-L | Ortho · ImageNet-1k classification (CA models) | Top-1 (%): 83.4 -> 84.2 | Improved |
| use UperNet instead of Semantic FPN decoder | Ortho-S · ADE20K semantic segmentation (CA encoders) | mIoU (%): 48.2 -> 48.5 | Improved |
| use UperNet instead of Semantic FPN decoder | Ortho-B · ADE20K semantic segmentation (CA encoders) | mIoU (%): 49.0 -> 49.8 | Improved |
| replace SimA with 8-token patching (T12/16 to T12/8) | XCiT SimA · ImageNet-1k classification (CA models) | Top-1 (%): 82.1 -> 79.4 | Hurt |
| scale from MViTv2-T to MViTv2-L | MViTv2 · COCO val2017 detection (CA encoders) | AP box: 48.2 -> 51.8 | Improved |
| scale from PVT-Tiny to PVT-Large | PVT · COCO val2017 detection (CA encoders) | AP box: 36.7 -> 42.9 | Improved |
Closest papers by meaning
- CvT: Introducing Convolutions to Vision Transformers (ICCV 2021 · similarity 0.93)
- Q-ViT: Accurate and Fully Quantized Low-bit Vision Transformer (NeurIPS 2022 · similarity 0.92)
- EfficientFormer: Vision Transformers at MobileNet Speed (NeurIPS 2022 · similarity 0.91)
- EfficientViT: Memory Efficient Vision Transformer With Cascaded Group Attention (CVPR 2023 · similarity 0.91)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.91)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.90)
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.
