Research map / Architectures and efficiency / Attention and transformers
Rethinking and Improving Relative Position Encoding for Vision Transformer
Kan Wu, Houwen Peng, Minghao Chen, Jianlong Fu, Hongyang Chao
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 |
|---|---|---|---|
| add learnable absolute position encoding to a model with no position encoding (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 77.6 -> 79.9 | Improved |
| replace absolute position encoding with relative encoding on keys (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 79.9 -> 80.9 | Improved |
| add absolute position encoding on top of relative encoding (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.9 -> 80.9 | No clear effect |
| apply relative encoding to values instead of queries or keys (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.9 -> 80.2 | Hurt |
| add relative encoding on queries and values to keys (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.9 -> 81.3 | Improved |
| use directed Product instead of undirected Euclidean mapping (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.4 -> 80.9 | Improved |
| use contextual instead of bias mode for Product method (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.5 -> 80.9 | Improved |
| share bias-mode encoding across attention heads (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.54 -> 80.05 | Hurt |
| share contextual-mode encoding across attention heads (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.99 -> 80.89 | No clear effect |
| use piecewise instead of clip index function on classification (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 80.1 -> 80.0 | No clear effect |
| use piecewise instead of clip index function (ablation) | DETR-ResNet50 · COCO val, 150 epochs | AP: 40.4 -> 40.8 | Improved |
| increase number of buckets up to 50 (ablation) | DeiT-S contextual Product · ImageNet-1k val | top-1 accuracy: 79.9 -> 80.9 | Improved |
Closest papers by meaning
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.90)
- Parabolic Position Encoding: Vision-Centric, Principled, Extrapolatable, General (NeurIPS 2026 · similarity 0.89)
- KeyPoint Relative Position Encoding for Face Recognition (CVPR 2024 · similarity 0.87)
- DAPE: Data-Adaptive Positional Encoding for Length Extrapolation (NeurIPS 2024 · similarity 0.87)
- Toward Relative Positional Encoding in Spiking Transformers (NeurIPS 2025 · similarity 0.87)
- Maximizing the Position Embedding for Vision Transformers with Global Average Pooling (AAAI 2025 · similarity 0.87)
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.
