Research map / Architectures and efficiency / Attention and transformers
ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases
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 |
|---|---|---|---|
| replace DeiT with ConViT-Ti | DeiT-Ti · ImageNet-1k test | Top-1 accuracy: 72.2 -> 73.1 | Helped |
| replace DeiT with ConViT-Ti+ | DeiT-Ti+ · ImageNet-1k test | Top-1 accuracy: 75.9 -> 76.7 | Helped |
| replace DeiT with ConViT-S | DeiT-S* · ImageNet-1k test | Top-1 accuracy: 79.8 -> 81.3 | Helped |
| replace DeiT with ConViT-S+ | DeiT-S+ · ImageNet-1k test | Top-1 accuracy: 79.0 -> 82.2 | Helped |
| replace DeiT with ConViT-B | DeiT-B* · ImageNet-1k test | Top-1 accuracy: 81.8 -> 82.4 | Helped |
| replace DeiT with ConViT-B+ | DeiT-B+ · ImageNet-1k test | Top-1 accuracy: 77.5 -> 82.5 | Helped |
| add gating (ablation) | ConViT with convolution and GPSA · ImageNet-1k, full data | Top-1 accuracy: 82.0 -> 82.2 | Helped |
| add gating (ablation) | ConViT with convolution and GPSA · ImageNet-1k, 10% data | Top-1 accuracy: 46.1 -> 49.4 | Helped |
| add convolution (ablation) | ConViT with gating and GPSA · ImageNet-1k, full data | Top-1 accuracy: 81.4 -> 82.2 | Helped |
| add convolution (ablation) | ConViT with gating and GPSA · ImageNet-1k, 10% data | Top-1 accuracy: 43.9 -> 49.4 | Helped |
| add both gating and convolution (ablation) | ConViT with GPSA · ImageNet-1k, full data | Top-1 accuracy: 81.6 -> 82.2 | Helped |
| add both gating and convolution (ablation) | ConViT with GPSA · ImageNet-1k, 10% data | Top-1 accuracy: 42.0 -> 49.4 | Helped |
Closest papers by meaning
- RegionViT: Regional-to-Local Attention for Vision Transformers (ICLR 2022 · similarity 0.91)
- Do Vision Transformers See Like Convolutional Neural Networks? (NeurIPS 2021 · similarity 0.90)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.90)
- Structured Initialization for Vision Transformers (NeurIPS 2025 · similarity 0.89)
- Skip-Attention: Improving Vision Transformers by Paying Less Attention (ICLR 2024 · similarity 0.89)
- FasterViT: Fast Vision Transformers with Hierarchical Attention (ICLR 2024 · similarity 0.89)
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.