Research map / Vision-language and multimodal / Human and embodied modeling
HSGG: Training-Free Hierarchical Scene Graph Generation with Geometry-Guided Relation Reasoning
yunzhe Liu, Wenbiao Liu, Lihui Cen, Zhe Qu, Yigang Cen, Yixiong Liang, Yongfang Xie, Shichao Kan
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 BHEP entity perception (vs GG-CRD alone) (ablation) | HSGG · PSG zero-shot SGG | zR@100: 7.35 -> 11.63 | Improved |
| add GA-RF relation filtering on top of BHEP (ablation) | HSGG · PSG zero-shot SGG | zR@100: 11.63 -> 12.13 | Improved |
| add GG-CRD contrastive decoding on top of BHEP + GA-RF (ablation) | HSGG · PSG zero-shot SGG | zR@100: 12.13 -> 14.05 | Improved |
| add GA-RF to GG-CRD without BHEP (ablation) | HSGG · PSG zero-shot SGG | zR@100: 7.35 -> 7.58 | Improved |
| use HSGG vs training-free OWSGG on VG zero-shot | zero-shot SGG · VG150 | R@50/100: +12.3% / +13.1% (R@50/100) | Improved |
| use HSGG vs trained CAPSGG on VG zero-shot | zero-shot SGG · VG150 | R@50/100: +1.0% / +1.8% | Improved |
| use HSGG vs OWSGG on PSG zero-shot | zero-shot SGG · PSG | R@50/100: +6.7% / +7.7% | Improved |
| use HSGG vs OvSGTR on open-world VG | open-world SGG · VG open-world | R@50/100: +3.3% / +1.9% | Improved |
| add HSG scene graphs to InternVL3-8B | InternVL3-8B · V* Bench | overall accuracy: 71.73 -> 76.43 | Improved |
| add HSG scene graphs to LLaVA-OV-1.5-8B | LLaVA-OV-1.5-8B · V* Bench | overall accuracy: 76.96 -> 79.58 | Improved |
| use HSG scene graph as sole input instead of image for captioning | LLaVA-OV-1.5-8B · CompreCap | Sunified: 59.70 -> 62.97 | Improved |
| use GG-CRD decoding instead of regular decoding | LLaVA-1.5-7B · relational hallucination benchmarks | average accuracy: +9.60% relative to regular decoding | Improved |
Closest papers by meaning
- Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models (ICML 2026 · similarity 0.89)
- RelationVGGT : Visual Geometry Transformers for 3D Spatial Relation Segmentation (NeurIPS 2026 · similarity 0.89)
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · similarity 0.89)
- HyperGLM: HyperGraph for Video Scene Graph Generation and Anticipation (CVPR 2025 · similarity 0.89)
- ChainMPQ: Interleaved Text-Image Reasoning Chains for Mitigating Relation Hallucinations (ICLR 2026 · similarity 0.88)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.88)
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.
