Research map / Label-efficient and robust learning / Recognition and long-tail
Comprehensive Attribution: Inherently Explainable Vision Model with Feature Detector
Xianren Zhang, Dongwon Lee, Suhang Wang
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 |
|---|---|---|---|
| pre-train the detector on full images and keep it fixed, instead of training it on masked-out regions (ablation) | COMET · ImageNet-9 | PxAP: 71.41 -> 76.38 | Improved |
| pre-train the detector on full images instead of training it on masked-out regions (ablation) | COMET · ImageNet-9 | accuracy (%): 92.82 -> 94.33 | Improved |
| pre-train the detector on full images instead of training it on masked-out regions (ablation) | COMET · NICO++ | accuracy (%): 58.38 -> 56.47 | Hurt |
| add the pre-trained detector term on masked-out regions to the selector objective | COMET · ImageNet-9 | classification accuracy: 91.16 -> 94.33 | Improved |
| add the detector (versus COMET w/o Detector, where predictor and selector interlock) (ablation) | COMET · ImageNet-9 | accuracy | Improved |
| set the masked-out penalty weight a to 0 (no control on masked-out regions) (ablation) | COMET · ImageNet-9 | PxAP | Hurt |
| use COMET instead of existing inherently explainable baselines for noisy NICO++ accuracy | COMET · NICO++ (noisy) | accuracy (%): 28.97 -> 33.12 | Improved |
| use COMET for localization against best post-hoc baseline | COMET · ImageNet-9 | PxAP: 60.67 -> 76.38 | Improved |
| swap the ResNet18 predictor for a ViT in COMET | COMET · ImageNet-9 | PxAP: 76.38 -> 69.87 | Hurt |
| train COMET on randomized labels (sanity check) (ablation) | COMET · an8Flower | area under IOU curve: 72.5 -> 5.83 | Hurt |
| use COMET for fidelity (lower perturbed accuracy) versus other inherent methods | COMET · ImageNet-9 (top 20% removed) | perturbed accuracy (%): 21.48 -> 2.07 | Improved |
Closest papers by meaning
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.91)
- How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations (ICLR 2025 · similarity 0.87)
- Less is More: Fewer Interpretable Region via Submodular Subset Selection (ICLR 2024 · similarity 0.87)
- GEFA: A General Feature Attribution Framework Using Proxy Gradient Estimation (ICML 2025 · similarity 0.87)
- One Wave To Explain Them All: A Unifying Perspective On Feature Attribution (ICML 2025 · similarity 0.87)
- Pixel-level Certified Explanations via Randomized Smoothing (ICML 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.
