Research map / Architectures and efficiency / Attention and transformers
Locality-Attending Vision Transformer
Sina Hajimiri, Farzad Beizaee, Fereshteh Shakeri, Christian Desrosiers, Ismail Ben Ayed, Jose Dolz
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 LocAt (GAug + PRR) to ViT | ViT-Tiny · ADE20K | mIoU: 17.30 -> 23.47 | Improved |
| add LocAt to ViT | ViT-Tiny · ImageNet-1K | top-1: 72.39 -> 73.94 | Improved |
| add LocAt to Swin | Swin-Tiny · ADE20K | mIoU: 25.58 -> 26.52 | Improved |
| add LocAt to Jumbo | Jumbo-Base · ImageNet-1K | top-1: 84.42 -> 84.43 | No clear effect |
| add LocAt to ViT trained on small datasets | ViT-Tiny · CIFAR-100 | top-1: 73.84 -> 80.43 | Improved |
| add LocAt to ViT in DINO self-supervised training | DINO ViT-S/16 · ImageNet-1K linear | top-1: 65.52 -> 67.65 | Improved |
| add GAug (Gaussian-augmented attention) to ViT (ablation) | ViT-Tiny · ADE20K | mIoU: 17.30 -> 18.98 | Improved |
| add PRR (patch representation refinement) to ViT (ablation) | ViT-Tiny · ADE20K | mIoU: 17.30 -> 21.60 | Improved |
| add PRR to ViT-GAug (full LocAt) (ablation) | ViT-Tiny · ADE20K | mIoU: 18.98 -> 23.47 | Improved |
| use PRR instead of GAP on ViT (ablation) | ViT-Base · ADE20K | mIoU: 27.99 -> 29.89 | Improved |
| add global average pooling to ViT (ablation) | ViT-Base · ADE20K | mIoU: 28.40 -> 27.99 | Hurt |
| add LocAt to ViT without positional embeddings (ablation) | ViT-Tiny · ADE20K | mIoU: 15.13 -> 22.69 | Improved |
Closest papers by meaning
- Vision Transformers: From Semantic Segmentation to Dense Prediction (IJCV 2024 · similarity 0.91)
- Vicinity Vision Transformer (TPAMI 2023 · similarity 0.91)
- RegionViT: Regional-to-Local Attention for Vision Transformers (ICLR 2022 · similarity 0.90)
- Less Is More: Pay Less Attention in Vision Transformers (AAAI 2022 · similarity 0.89)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · similarity 0.88)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · 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.
