Research map / Label-efficient and robust learning / Recognition and long-tail
Hidden in Plain Sight -- Class Competition Focuses Attribution Maps
Nils Philipp Walter, Jilles Vreeken, 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 |
|---|---|---|---|
| add Attribution Lens (class-competition softmax over attribution maps) | ResNet-50 attribution methods · Quad-ImageNet | RA: avg +53.17% | Improved |
| add Attribution Lens to Guided Backprop | GBP on ResNet-50 · Quad-ImageNet | RA: +144% | Improved |
| add Attribution Lens to ViT attribution methods | ViT-base-16 methods · Quad-ImageNet | IoU: avg +103.29% | Improved |
| add Attribution Lens to Input x Gradient and Integrated Gradients | IxG, IG on ResNet-50 · Quad-ImageNet | IoU: 0.20, +0% | No clear effect |
| add Attribution Lens on real multi-object images | ResNet-50 methods · COCO | RA, IoU, F1: avg +17.83, +5.83, +4.83 | Improved |
| add Attribution Lens on CNN insertion test | CNN attribution methods · Insertion AUC | AUC: 8-11% on average | Improved |
| add Attribution Lens on transformer insertion test | ViT attribution methods · Insertion AUC | AUC: ~2% | Improved |
| add Attribution Lens to Grad-CAM | GradCam · Insertion AUC | AUC: +0% -> -6% (ConvNeXT) | Mixed |
| add Attribution Lens to Grad-CAM on DenseNet121 | GradCam DenseNet121 · COCO | IoU: -25% | Hurt |
| add Attribution Lens on ConvNeXt | ConvNeXt methods · Quad-ImageNet | IoU: avg -1.20% | Hurt |
| add Attribution Lens to perturbation-based methods | Shapley, LIME, Occlusion etc. · COCO | RA vs IoU/F1: RA +13 but IoU -1% | Mixed |
| add Attribution Lens before randomization sanity check | GBP, IxG, ViT methods · Cascading randomization | Spearman similarity | Improved |
Closest papers by meaning
- Attribution Explanations for Deep Neural Networks: A Theoretical Perspective (TPAMI 2026 · similarity 0.89)
- A Unified Taylor Framework for Revisiting Attribution Methods (AAAI 2021 · similarity 0.88)
- Evaluating Attribution for Graph Neural Networks (NeurIPS 2020 · similarity 0.88)
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.87)
- One Wave To Explain Them All: A Unifying Perspective On Feature Attribution (ICML 2025 · similarity 0.86)
- How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations (ICLR 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.
