Research map / Architectures and efficiency / Attention and transformers
Spatial Priors via Space Filling Curves for Small and Limited Data Vision Transformers
Leyla Naz Candogan, Arshia Afzal, Pol Puigdemont, Volkan Cevher
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 Violin SFC decay mask at fine-tuning | DeiT-T · VTAB-1K full fine-tuning | avg top-1 accuracy (%): 65.52 -> 68.33 | Improved |
| add Violin SFC decay mask at fine-tuning | DeiT-III-S · VTAB-1K Structured | accuracy (%): 52.92 -> 61.61 | Improved |
| add Violin mask at fine-tuning | DINO-S · VTAB-1K full fine-tuning | avg top-1 accuracy (%): 71.21 -> 71.84 | Improved |
| add Violin mask at fine-tuning | DeiT-III-H · VTAB-1K full fine-tuning | avg top-1 accuracy (%): 66.91 -> 68.50 | Improved |
| add Violin mask on top of LoRA | LoRA · VTAB-1K PEFT | avg accuracy (%): 71.04 -> 72.55 | Improved |
| add Violin mask on top of DoRA | DoRA · VTAB-1K PEFT | avg accuracy (%): 70.75 -> 71.90 | Improved |
| add Violin to pixel-level ViT | DeiT-T · Pixel-level CIFAR-100 | avg accuracy (%): 60.8 -> 68.0 | Improved |
| add Violin during ImageNet pretraining | DeiT-T · ImageNet-1K | top-1 accuracy: +0.8% | Improved |
| add Violin to Swin-T Mask R-CNN | Swin-T · COCO detection | mAP: +0.4 | Improved |
| use four-curve Violin instead of a single Peano curve (ablation) | Violin · VTAB-1K Structured | avg accuracy (%): 61.63 -> 61.89 | Improved |
| use multiplicative averaged mask instead of additive (ablation) | Violin · VTAB-1K Structured | avg accuracy (%): 61.34 -> 61.89 | Improved |
| use SFC masks instead of Manhattan distance mask (ablation) | Violin · VTAB-1K Structured | avg accuracy (%): 58.37 -> 61.89 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DoRA — DoRA: Weight-Decomposed Low-Rank Adaptation (ICML 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.87)
- LookHere: Vision Transformers with Directed Attention Generalize and Extrapolate (NeurIPS 2024 · similarity 0.86)
- LF-ViT: Reducing Spatial Redundancy in Vision Transformer for Efficient Image Recognition (AAAI 2024 · similarity 0.85)
- Breaking the Low-Rank Dilemma of Linear Attention (CVPR 2025 · similarity 0.85)
- SATA: Spatial Autocorrelation Token Analysis for Enhancing the Robustness of Vision Transformers (CVPR 2025 · similarity 0.85)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 2024 · similarity 0.85)
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.
