Research map / Architectures and efficiency / Attention and transformers
VOLO: Vision Outlooker for Visual Recognition
Li Yuan, Qibin Hou, Zihang Jiang, Jiashi Feng, Shuicheng Yan
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 |
|---|---|---|---|
| replace 2 transformer layers with Outlookers (ablation) | LV-ViT-S · ImageNet-1k val | top-1 accuracy: 83.3 -> 83.7 | Improved |
| add 2 more Outlookers (ablation) | LV-ViT-S with 2 Outlookers · ImageNet-1k val | top-1 accuracy: 83.7 -> 84.0 | Improved |
| increase transformer heads from 6 to 12 (ablation) | VOLO-D1 · ImageNet-1k val | top-1 accuracy: 84.0 -> 84.2 | Improved |
| raise test resolution from 224 to 384 by finetuning (ablation) | VOLO-D1 · ImageNet-1k val | top-1 accuracy: 84.2 -> 85.2 | Improved |
| use Outlooker instead of local self-attention (ablation) | VOLO-D1 · ImageNet-1k val | top-1 accuracy: 83.8 -> 84.2 | Improved |
| use Outlooker instead of 3x3 convolution (ablation) | VOLO-D1 · ImageNet-1k val | top-1 accuracy: 83.8 -> 84.2 | Improved |
| increase Outlookers from 4 to 6 (ablation) | VOLO-D1 · ImageNet-1k val | top-1 accuracy: 84.0 -> 83.9 | No clear effect |
| increase Outlooker heads from 2 to 6 (ablation) | VOLO-D1 · ImageNet-1k val | top-1 accuracy: 83.9 -> 84.0 | Improved |
| increase Outlooker heads from 6 to 8 (ablation) | VOLO-D1 · ImageNet-1k val | top-1 accuracy: 84.0 -> 84.0 | No clear effect |
| scale from VOLO-D1 to VOLO-D2 | VOLO · ImageNet-1k val | top-1 accuracy: 84.2 -> 85.2 | Improved |
| scale from VOLO-D2 to VOLO-D5 | VOLO · ImageNet-1k val | top-1 accuracy: 85.2 -> 86.1 | Improved |
| add Outlookers on 8x8 patch tokens (any number) over no Outlookers (ablation) | LV-ViT-S · ImageNet-1k val | top-1 accuracy: 83.3 -> 84.0 | Improved |
Closest papers by meaning
- UniFormer: Unifying Convolution and Self-Attention for Visual Recognition (TPAMI 2023 · similarity 0.89)
- Involution: Inverting the Inherence of Convolution for Visual Recognition (CVPR 2021 · similarity 0.88)
- Learn to Pay Attention (ICLR 2018 · similarity 0.88)
- Conv2Former: A Simple Transformer-Style ConvNet for Visual Recognition (TPAMI 2024 · similarity 0.88)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.87)
- Linear Differential Vision Transformer: Learning Visual Contrasts via Pairwise Differentials (NeurIPS 2025 · similarity 0.86)
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.
