Research map / Architectures and efficiency / CNN design
There and Back Again: Revisiting Backpropagation Saliency Methods
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 |
|---|---|---|---|
| Combine saliency maps from multiple layers | Linear approximation (LA) · Weak localization evaluations | localization performance | Helped |
| Combine saliency maps from multiple layers | Selective NormGrad (sNG and MsNG) · Weak localization evaluations | localization performance | Helped |
| Combine saliency maps from multiple layers | Grad-CAM and other methods without useful information at earlier layers · Weak localization evaluations | localization performance | No clear effect |
| Add the test-time meta-saliency inner SGD step | Backpropagation saliency methods · Class-sensitivity evaluations | class sensitivity | Helped |
| Use selective rather than non-selective NormGrad | NormGrad variants · All 8 evaluations tested | evaluation performance | Helped |
| Use selective NormGrad rather than excitation backprop | Saliency methods · All 8 evaluations tested | evaluation performance | Helped |
| Use selective NormGrad rather than gradients sum | Saliency methods · All 8 evaluations tested | evaluation performance | Helped |
| Use selective NormGrad rather than guided backprop | Saliency methods · All 8 evaluations tested | evaluation performance | Helped |
| Use output rather than input activations when forming saliency | Unified saliency formulation · Saliency maps | map difference | No clear effect |
Closest papers by meaning
- CAMERAS: Enhanced Resolution and Sanity Preserving Class Activation Mapping for Image Saliency (CVPR 2021 · similarity 0.92)
- Understanding and Visualizing Deep Visual Saliency Models (CVPR 2019 · similarity 0.89)
- Where do Models go Wrong? Parameter-Space Saliency Maps for Explainability (NeurIPS 2022 · similarity 0.89)
- Sanity Checks for Saliency Maps (NeurIPS 2018 · similarity 0.89)
- Shallow and Deep Convolutional Networks for Saliency Prediction (CVPR 2016 · similarity 0.88)
- Explainable Saliency: Articulating Reasoning with Contextual Prioritization (CVPR 2025 · similarity 0.87)
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.