Research map / Image generation / Autoregressive and latent generation
XRAI: Better Attributions Through Regions
Andrei Kapishnikov, Tolga Bolukbasi, Fernanda Viegas, Michael Terry
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 black+white baselines instead of a black baseline for integrated gradients | Integrated Gradients · ImageNet val | area under SIC (Resnet50-V2): 0.515 -> 0.575 | Improved |
| use black+white baselines instead of a black baseline for integrated gradients | Integrated Gradients · ImageNet segmentation dataset | AUC: 0.710 -> 0.729 | Improved |
| use 4 random baselines instead of a black baseline for integrated gradients | Integrated Gradients · ImageNet val | area under SIC (Resnet50-V2): 0.515 -> 0.623 | Improved |
| use 4 random baselines instead of black+white baselines for integrated gradients | Integrated Gradients · ImageNet val | area under SIC (Inception): 0.601 -> 0.595 | No clear effect |
| add region-level aggregation of IG attributions (XRAI) on top of IG B+W | IG (B+W) · ImageNet val | area under SIC (Resnet50-V2): 0.575 -> 0.749 | Improved |
| add region-level aggregation of IG attributions (XRAI) on top of IG B+W | IG (B+W) · ImageNet val | area under AIC (Inception): 0.634 -> 0.727 | Improved |
| use XRAI instead of GradCam | saliency method · ImageNet val | area under SIC (Inception): 0.703 -> 0.720 | Improved |
| use XRAI instead of GradCam | saliency method · ImageNet val | area under SIC (Resnet50-V2): 0.760 -> 0.749 | Hurt |
| use XRAI instead of GradCam | saliency method · ImageNet val | area under AIC (Resnet50-V2): 0.727 -> 0.728 | No clear effect |
| use XRAI instead of GradCAM | saliency method · ImageNet segmentation dataset | AUC: 0.742 -> 0.836 | Improved |
| use XRAI instead of GradCAM | saliency method · ImageNet segmentation dataset | MAE: 0.194 -> 0.149 | Improved |
| use XRAI instead of GradCAM | saliency method · ImageNet segmentation dataset | F1: 0.715 -> 0.786 | Improved |
Closest papers by meaning
- Towards Better Understanding Attribution Methods (CVPR 2022 · similarity 0.88)
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.88)
- Better Understanding Differences in Attribution Methods via Systematic Evaluations (TPAMI 2024 · similarity 0.88)
- Sanity Checks for Saliency Maps (NeurIPS 2018 · similarity 0.86)
- ADD: Attribution-Driven Data Augmentation Framework for Boosting Image Super-Resolution (CVPR 2025 · similarity 0.86)
- One Wave To Explain Them All: A Unifying Perspective On Feature Attribution (ICML 2025 · similarity 0.85)
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.
