Research map / Graph learning / Knowledge and temporal graphs
Explainable and Explicit Visual Reasoning Over Scene Graphs
Jiaxin Shi, Hanwang Zhang, Juanzi Li
AI-extracted. The measured effects below were read by AI from the paper's full text (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 ground-truth scene graphs instead of detected scene graphs | XNM · CLEVR validation | overall accuracy: 97.8 -> 99.9 | Improved |
| Use ground-truth scene graphs instead of detected scene graphs | XNM · CLEVR-CoGenT Condition B validation | accuracy: 72.1 -> 99.9 | Improved |
| Use ground-truth scene graphs rather than detected RoI features | XNM · CLEVR-CoGenT Condition B | accuracy: 72.1 -> 99.9 | Improved |
| Replace ground-truth programs with parsed programs | XNM-GT · CLEVR validation | overall accuracy: 100 -> 99.9 | Hurt |
| Replace ground-truth programs with parsed programs | XNM-Det · CLEVR validation | overall accuracy: 97.9 -> 97.8 | Hurt |
| Use scene-graph reasoning instead of the Up-Down baseline | VQAv2.0 · validation | accuracy: 63.2 -> 64.7 | Improved |
| Use scene-graph reasoning instead of StackNMN | VQAv2.0 · test | accuracy: 64.1 -> 67.5 | Improved |
| Use ground-truth scene graphs and programs | XNM-GT · CLEVR validation | overall accuracy: 99.9 -> 100 | Improved |
| Use ground-truth scene graphs and programs | XNM-GT · CLEVR-CoGenT Condition B validation | accuracy: 99.9 -> 100 | Improved |
| Train with only 10% of the training data | XNM-GT · CLEVR validation | accuracy: 100% | No clear effect |
| Use sigmoid attention instead of label-space softmax (ablation) | XNM-GT · CLEVR validation | accuracy: 100% | No clear effect |
Closest papers by meaning
- Explicit Knowledge Incorporation for Visual Reasoning (CVPR 2021 · similarity 0.90)
- Transparency by Design: Closing the Gap Between Performance and Interpretability in Visual Reasoning (CVPR 2018 · similarity 0.89)
- Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models (ICML 2026 · similarity 0.88)
- Visual Graph Arena: Evaluating Visual Conceptualization of Vision and Multimodal Large Language Models (ICML 2025 · similarity 0.87)
- VisionGraph: Leveraging Large Multimodal Models for Graph Theory Problems in Visual Context (ICML 2024 · similarity 0.86)
- Synthetic Visual Genome (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.
