Research map / Architectures and efficiency / Attention and transformers
Adversarial Normalization: I Can Visualize Everything (ICE)
Hoyoung Choi, Seungwan Jin, Kyungsik Han
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 patch-wise classification with adversarial normalization | DeiT-S · ImageNet-Segmentation | Pixel-wise accuracy: 4.05% | Improved |
| Add patch-wise classification with adversarial normalization | DeiT-S · ImageNet-Segmentation | Mean IoU: 3.94% | Improved |
| Add patch-wise classification with adversarial normalization | DeiT-S · PascalVOC12 | Jaccard similarity: 41.32 | Improved |
| Use ICE instead of DINO attention for foreground/background separation | DeiT-S · PascalVOC12 | Jaccard similarity: 40.30 -> 41.32 | Improved |
| Add patch-wise classification with adversarial normalization | DeiT-S · CUB-200-2011 | GT Loc: 92.5 | Improved |
| Add patch-wise classification with adversarial normalization | DeiT-S · PascalVOC07 and PascalVOC12 | CorLoc: 62.2; 66.0 | Mixed |
| Use ICE for patch selection instead of EViT | DeiT-S, keep-rate 0.5 · ImageNet | Top-1 accuracy (%): 78.39 -> 78.49 | Improved |
| Use ICE for patch selection instead of EViT | DeiT-S, keep-rate 0.5 · ImageNet | Throughput: 1701 -> 1547 | Hurt |
| Freeze DeiT-S parameters and train only the two MLP layers (ablation) | ICE · ImageNet-Segmentation | Mean IoU: 67.94 -> 62.12 | Hurt |
| Add background scale α=0.5 rather than use α=1.0 (ablation) | ICE with DeiT-S · ImageNet-Segmentation | Pixel-wise accuracy: 63.45 -> 84.85 | Improved |
| Add background scale α=0.5 rather than use α=1.0 (ablation) | ICE with DeiT-S · ImageNet-Segmentation | Mean IoU: 45.36 -> 67.94 | Improved |
| Use class weight λclass=0.975 rather than λclass=0.5 (ablation) | ICE with DeiT-S · ImageNet-Segmentation | Mean IoU: 57.59 -> 67.94 | Improved |
Closest papers by meaning
- Attention Guided CAM: Visual Explanations of Vision Transformer Guided by Self-Attention (AAAI 2024 · similarity 0.88)
- Vision Transformers Need Registers (ICLR 2024 · similarity 0.88)
- Probabilistic Conceptual Explainers: Trustworthy Conceptual Explanations for Vision Foundation Models (ICML 2024 · similarity 0.87)
- ICE: Intrinsic Concept Extraction from a Single Image via Diffusion Models (CVPR 2025 · similarity 0.87)
- LookHere: Vision Transformers with Directed Attention Generalize and Extrapolate (NeurIPS 2024 · similarity 0.87)
- Token Transformation Matters: Towards Faithful Post-hoc Explanation for Vision Transformer (CVPR 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.
