Research map / Label-efficient and robust learning / Recognition and long-tail
VITAL: More Understandable Feature Visualization through Distribution Alignment and Relevant Information Flow
Ada Görgün, Bernt Schiele, Jonas Fischer
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 |
|---|---|---|---|
| replace activation maximization with feature distribution matching (VITAL) vs MACO | VITAL · ImageNet class visualizations, ResNet50 | CLIP zero-shot Top-1: 12.87 -> 66.62 | Improved |
| VITAL vs Fourier baseline | VITAL · ImageNet class visualizations, ResNet50 | accuracy: 21.30 -> 99.90 | Improved |
| VITAL vs MACO accuracy on ViT-L-16 | VITAL · ImageNet class visualizations, ViT-L-16 | accuracy: 44.33 -> 99.80 | Improved |
| VITAL vs DeepInversion on FID (ResNet50) | VITAL · ImageNet class visualizations, ResNet50 | FID: 35.76 -> 58.79 | Hurt |
| VITAL vs MACO on FID (ConvNeXt) | VITAL · ImageNet class visualizations, ConvNeXt | FID: 62.55 -> 3.92 | Improved |
| VITAL vs MACO on FID (DenseNet121) | VITAL · ImageNet class visualizations, DenseNet121 | FID: 1.80 -> 0.27 | Improved |
| VITAL vs MACO on FID (ViT-L-16) | VITAL · ImageNet class visualizations, ViT-L-16 | FID: 946.96 -> 126.29 | Improved |
| VITAL vs DeepInversion on CLIP zero-shot Top-1 (DenseNet121) | VITAL · ImageNet class visualizations, DenseNet121 | CLIP zero-shot Top-1: 10.00 -> 58.70 | Improved |
| VITAL on ViT-L-32 reaches lower accuracy than on other architectures | VITAL · ImageNet class visualizations, ViT-L-32 | accuracy: 99.90 -> 89.60 | Mixed |
Closest papers by meaning
- Don't trust your eyes: on the (un)reliability of feature visualizations (ICML 2024 · similarity 0.87)
- Unlocking Feature Visualization for Deep Network with MAgnitude Constrained Optimization (NeurIPS 2023 · similarity 0.87)
- Understanding Visual Feature Reliance through the Lens of Complexity (NeurIPS 2024 · similarity 0.85)
- Manipulating Feature Visualizations with Gradient Slingshots (NeurIPS 2025 · similarity 0.83)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.83)
- Representational Similarity via Interpretable Visual Concepts (ICLR 2025 · similarity 0.83)
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.
