Research map / Vision-language and multimodal / Human and embodied modeling
Interpretable and Fine-Grained Visual Explanations for Convolutional Neural Networks
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 |
|---|---|---|---|
| Use FGVis instead of Sliding Window for pixel importance | ResNet50 · ImageNet validation | deletion metric: 0.1421 -> 0.0644 | Helped |
| Use FGVis instead of Sliding Window for pixel importance | VGG16 · ImageNet validation | deletion metric: 0.1158 -> 0.0636 | Helped |
| Apply neuron-activation gradient filtering during explanation optimization | FGVis · ImageNet validation images, VGG16 | adversarial class generated (%): 100.0\,\% -> 0.2\,\% | Helped |
| Apply neuron-activation gradient filtering during explanation optimization | FGVis · ImageNet validation images, AlexNet | adversarial class generated (%): 100.0\,\% -> 0.0\,\% | Helped |
| Apply neuron-activation gradient filtering during explanation optimization | FGVis · ImageNet validation images, ResNet50 | adversarial class generated (%): 100.0\,\% -> 0.0\,\% | Helped |
| Apply neuron-activation gradient filtering during explanation optimization | FGVis · ImageNet validation images, GoogleNet | adversarial class generated (%): 100.0\,\% -> 0.0\,\% | Helped |
| Use 50% rather than 25% overlap to count a lesion as detected (ablation) | FGVis · DiaretDB1, hemorrhages | image-level sensitivity (%): 100 -> 90.5 | Hurt |
| Use 50% rather than 25% overlap to count a lesion as detected (ablation) | FGVis · DiaretDB1, hard exudates | image-level sensitivity (%): 94.7 -> 81.6 | Hurt |
| Use 50% rather than 25% overlap to count a lesion as detected (ablation) | FGVis · DiaretDB1, soft exudates | image-level sensitivity (%): 90.0 -> 80.0 | Hurt |
| Use 50% rather than 25% overlap to count a lesion as detected (ablation) | FGVis · DiaretDB1, red small dots | image-level sensitivity (%): 88.4 -> 86.0 | Hurt |
| Use a zero reference image instead of Gaussian noise or blurred-image references (ablation) | FGVis · ImageNet reference-image entropy | entropy: Zero image: AlexNet 6.90, GoogleNet 4.08, VGG16 6.31, ResNet50 5.09 | Mixed |
Closest papers by meaning
- Comparing the Decision-Making Mechanisms by Transformers and CNNs via Explanation Methods (CVPR 2024 · similarity 0.87)
- Towards Fine-Grained Interpretability: Counterfactual Explanations for Misclassification with Saliency Partition (CVPR 2025 · similarity 0.87)
- How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations (ICLR 2025 · similarity 0.87)
- Keep the Faith: Faithful Explanations in Convolutional Neural Networks for Case-Based Reasoning (AAAI 2024 · similarity 0.86)
- Empowering CAM-Based Methods with Capability to Generate Fine-Grained and High-Faithfulness Explanations (AAAI 2024 · similarity 0.86)
- DiG-IN: Diffusion Guidance for Investigating Networks - Uncovering Classifier Differences Neuron Visualisations and Visual Counterfactual Explanations (CVPR 2024 · 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.