Research map / Detection and segmentation / CNN features
Grid Saliency for Context Explanations of Semantic Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| Use perturbation-based grid saliency instead of VG, IG, or SG | U-Net segmentation networks on synthetic context-biased data · Synthetic context-bias localization | mCBL: ~0.5 to above 0.9 | Helped |
| Use VG or SG context saliency instead of perturbation-based grid saliency | U-Net segmentation networks on synthetic context-biased data · Synthetic context-bias detection | mCBD | Hurt |
| Use IG instead of perturbation-based grid saliency | U-Net segmentation networks on synthetic context-biased data · Synthetic context-bias detection | mCBD | Mixed |
| Use VG, IG, or SG instead of perturbation-based grid saliency | U-Net segmentation networks on synthetic context-biased data · Synthetic context-bias localization | mCBL: ~0.5 to above 0.9 | Hurt |
| Increase the mask-loss weight λ (ablation) | Grid saliency · Synthetic context-bias detection and localization | mCBD and mCBL | Mixed |
| Vary the learning rate (ablation) | Grid saliency · Synthetic context-bias detection and localization | mCBD and mCBL | No clear effect |
| Vary the saliency-mask initialization (ablation) | Grid saliency · Synthetic context-bias detection and localization | mCBD and mCBL | No clear effect |
| Use U-Net with VGG16 rather than MobileNetv2 or ResNet18 (ablation) | Segmentation networks trained on weakly biased synthetic data · Unbiased synthetic test set | mIoU | Helped |
| Add the request mask to the perturbation formulation to isolate predictions in a target spatial region | Saliency for dense prediction networks · Semantic segmentation context explanations | spatial coherence | Helped |
| Clip mask activations at or below 0.2 to zero during optimization (ablation) | Grid saliency · Cityscapes context explanations | saliency-map noise | Helped |
Closest papers by meaning
- Explainable Saliency: Articulating Reasoning with Contextual Prioritization (CVPR 2025 · similarity 0.88)
- Visualization of Supervised and Self-Supervised Neural Networks via Attribution Guided Factorization (AAAI 2021 · similarity 0.87)
- Where do Models go Wrong? Parameter-Space Saliency Maps for Explainability (NeurIPS 2022 · similarity 0.87)
- There and Back Again: Revisiting Backpropagation Saliency Methods (CVPR 2020 · similarity 0.86)
- Sanity Checks for Saliency Maps (NeurIPS 2018 · similarity 0.86)
- Show and Tell: Visually Explainable Deep Neural Nets via Spatially-Aware Concept Bottleneck Models (CVPR 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.