Research map / Architectures and efficiency / Attention and transformers
DAVE: Distribution-Aware Attribution via ViT Gradient Decomposition
Adam Wróbel, Siddhartha Gairola, Jacek Tabor, Bernt Schiele, Bartosz Michał Zieliński, Dawid Damian Rymarczyk
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 |
|---|---|---|---|
| use DAVE instead of best competing method | ViT-B/16 · ImageNet-1k localization | GridPG (%): +1.79 | Improved |
| use DAVE instead of best competing method | DeiT-B/16 · ImageNet-1k localization | GridPG (%): +8.05 | Improved |
| use DAVE instead of best competing method | DeiT-III-B/16 · ImageNet-1k localization | GridPG (%): +11.92 | Improved |
| use DAVE instead of best competing method | DINO ViT-B/16 · ImageNet-1k localization | GridPG (%): +1.35 | Improved |
| use DAVE instead of best competing method (C-LRP) | ViT-B/16 · ImageNet-1k localization | EnergyPG (%): 80.82 -> 78.60 | Hurt |
| use DAVE instead of best competing method | DINO ViT-B/16 · ImageNet-1k localization | EnergyPG (%): +1.12 | Improved |
| use DAVE instead of inherent B-cos explanation | B-cos-ViT · ImageNet-1k localization (B-cos models) | GridPG (%): 79.67 -> 84.00 | Improved |
| use DAVE instead of inherent B-cos explanation | B-cos-ViT · ImageNet-1k localization (B-cos models) | EnergyPG (%): 69.41 -> 78.55 | Improved |
| use DAVE instead of inherent B-cos explanation | B-cos-ViT-C · ImageNet-1k localization (B-cos models) | GridPG (%): 87.66 -> 88.43 | Improved |
| use DAVE instead of inherent B-cos explanation | B-cos-ViT-C · ImageNet-1k localization (B-cos models) | EnergyPG (%): 75.61 -> 79.63 | Improved |
| increase neighborhood samples beyond 50 | DAVE · ImageNet-1k localization | GridPG/EnergyPG | No clear effect |
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.
- LeGrad — LeGrad: An Explainability Method for Vision Transformers via Feature Formation Sensitivity (ICCV 2025) · better on 8 of 8 shared results
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
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.89)
- LibraGrad: Balancing Gradient Flow for Universally Better Vision Transformer Attributions (CVPR 2025 · similarity 0.88)
- Vision Transformers Need Registers (ICLR 2024 · similarity 0.87)
- Pixel-level Certified Explanations via Randomized Smoothing (ICML 2025 · similarity 0.87)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · similarity 0.86)
- Vision Transformers with Self-Distilled Registers (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.
