Research map / Vision-language and multimodal / Visual question answering
Knowledge-Based Embodied Question Answering
Sinan Tan, Mengmeng Ge, Di Guo, Huaping Liu, Fuchun Sun
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add the knowledge graph to the program synthesizer (ablation) | Ours · K-EQA test | overall accuracy: 31.8 -> 41.3 | Improved |
| add planner-selected relevant-region exploration instead of full scan (ablation) | Ours · K-EQA test | overall path length: 131.1 -> 82.7 | Improved |
| add planner-selected region exploration instead of full scan (ablation) | Ours · K-EQA test | overall accuracy: 42.0 -> 41.3 | No clear effect |
| replace predicted segmentation with ground truth segmentation (ablation) | Ours · K-EQA test | overall accuracy: 41.3 -> 67.5 | Improved |
| replace voxel scene graph with ground truth scene graph (ablation) | Ours · K-EQA test | overall accuracy: 41.3 -> 90.3 | Improved |
| replace predicted segmentation with ground truth under full scan (ablation) | Ours + FS · K-EQA test | overall accuracy: 42.0 -> 70.0 | Improved |
| evaluate on K-EQA Extension with logic connectors instead of K-EQA | Ours · K-EQA vs Extension | overall accuracy: 41.3 -> 37.1 | Hurt |
| increase agents from 1 to 5 | Ours · K-EQA multi-agent | speedup: 1.00 -> 2.32 | Improved |
| increase agents from 1 to 5 | Ours · K-EQA multi-agent | accuracy: 41.3 -> 40.4 | No clear effect |
| increase turns from 1 to 10 with scene graph memory | Ours · K-EQA multi-turn | path length: 82.7 -> 12.5 | Improved |
| increase turns from 1 to 10 with scene graph memory | Ours · K-EQA multi-turn | accuracy: 41.3 -> 42.7 | No clear effect |
Closest papers by meaning
- Multi-Target Embodied Question Answering (CVPR 2019 · similarity 0.93)
- Embodied Question Answering (CVPR 2018 · similarity 0.91)
- Extending Embodied Question Answering from Perception to Decision (CVPR 2026 · similarity 0.91)
- OpenEQA: Embodied Question Answering in the Era of Foundation Models (CVPR 2024 · similarity 0.90)
- Multi-modal Situated Reasoning in 3D Scenes (NeurIPS 2024 · similarity 0.89)
- IndustryEQA: Pushing the Frontiers of Embodied Question Answering in Industrial Scenarios (NeurIPS 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.
