Research map / Detection and segmentation / CNN features
ODAM: Gradient-based Instance-Specific Visual Explanations for Object Detection
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 instance-specific gradient maps instead of perturbation maps | ODAM explanations · MS COCO | PG(b) accuracy: 82.6 -> 91.9 | Helped |
| Use instance-specific gradient maps instead of perturbation maps | ODAM explanations · CrowdHuman | PG(b) accuracy: 1.5 -> 95.5 | Helped |
| Add Odam-Train consistency and separation losses (ablation) | ODAM explanations · MS COCO | PG(b) accuracy: 91.9 -> 93.3 | Helped |
| Add Odam-Train consistency and separation losses (ablation) | ODAM explanations · CrowdHuman | PG(b) accuracy: 95.5 -> 97.3 | Helped |
| Add Odam-Train consistency and separation losses (ablation) | FCOS with Odam-NMS · CrowdHuman | JI: 78.5 -> 81.1 | Helped |
| Add Odam-Train consistency and separation losses (ablation) | Faster R-CNN with Odam-NMS · CrowdHuman | JI: 78.7 -> 80.5 | Helped |
| Replace NMS with Odam-NMS using heat-map correlations | FCOS · CrowdHuman | JI: 78.4 -> 81.1 | Helped |
| Replace NMS with Odam-NMS using heat-map correlations | Faster R-CNN · CrowdHuman | AP: 86.9 -> 88.1 | Helped |
| Add Odam-Train to detector training while retaining classical NMS (ablation) | FCOS · CrowdHuman | AP: 87.8 -> 87.5 | No clear effect |
| Add Odam-Train to detector training while retaining classical NMS (ablation) | Faster R-CNN · CrowdHuman | AP: 86.9 -> 87.0 | No clear effect |
| Use Odam-Train for faithfulness-oriented explanations (ablation) | ODAM · MS COCO | faithfulness | Hurt |
Closest papers by meaning
- Black-Box Explanation of Object Detectors via Saliency Maps (CVPR 2021 · similarity 0.89)
- FFAM: Feature Factorization Activation Map for Explanation of 3D Detectors (NeurIPS 2024 · similarity 0.87)
- Distilling Object Detectors With Fine-Grained Feature Imitation (CVPR 2019 · similarity 0.87)
- Interpreting Object-level Foundation Models via Visual Precision Search (CVPR 2025 · similarity 0.86)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.85)
- Training-Free Open-Ended Object Detection and Segmentation via Attention as Prompts (NeurIPS 2024 · 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.