Research map / Image generation / Image restoration
Guided Integrated Gradients: An Adaptive Path Method for Removing Noise
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Adapt the integration path by selecting low-absolute-gradient pixels (GIG(0)) instead of using a straight path | Integrated Gradients · ImageNet-Inception, black baseline | AUC-ROC: 0.655 -> 0.712 | Helped |
| Adapt the integration path with GIG(0) instead of using a straight path | Integrated Gradients · Open Images, black baseline | AUC-ROC: 0.557 -> 0.630 | Helped |
| Use unbounded GIG(0) rather than straight-path IG | Integrated Gradients · DR, black baseline | AUC-ROC: 0.833 -> 0.619 | Hurt |
| Add 40 anchors to Guided IG (ablation) | Guided IG · DR, black baseline | AUC-ROC: 0.619 -> 0.865 | Helped |
| Add 40 anchors to Guided IG (ablation) | Guided IG · ImageNet-Inception, black baseline | AUC-ROC: 0.712 -> 0.690 | Hurt |
| Combine SmoothGrad with Guided IG instead of SmoothGrad with IG | SmoothGrad attribution · ImageNet-Inception | AUC-ROC: 0.773 -> 0.776 | No clear effect |
| Combine XRAI with Guided IG instead of XRAI with IG | XRAI · ImageNet-Inception, black baseline | AUC-ROC: 0.765 -> 0.829 | Helped |
| Combine XRAI with Guided IG instead of XRAI with IG | XRAI · DR, black baseline | AUC-ROC: 0.793 -> 0.630 | Hurt |
| Use GIG(40) instead of GIG(0) (ablation) | Guided IG · ImageNet, MobileNet | SIC AUC: 0.423 -> 0.453 | Helped |
| Use a black-and-white baseline rather than a black baseline (ablation) | GIG(0) · ImageNet-Inception | AUC-ROC: 0.712 -> 0.738 | Helped |
| Use Guided IG rather than IG for closed-path attribution | Attribution methods · ImageNet validation, MobileNet | closed path mean squared error: 3.817e-07 -> 7.320e-08 | Helped |
Closest papers by meaning
- Integrated Decision Gradients: Compute Your Attributions Where the Model Makes Its Decision (AAAI 2024 · similarity 0.90)
- Denoising Diffusion Path: Attribution Noise Reduction with An Auxiliary Diffusion Model (NeurIPS 2024 · similarity 0.89)
- Manifold Integrated Gradients: Riemannian Geometry for Feature Attribution (ICML 2024 · similarity 0.86)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.84)
- Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models (AAAI 2025 · similarity 0.84)
- Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance (CVPR 2024 · similarity 0.84)
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.