AI Paper Atlas

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

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Scene graph generation

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 changedWhereMeasuredEffect
add BHEP entity perception (vs GG-CRD alone) (ablation)HSGG · PSG zero-shot SGGzR@100: 7.35 -> 11.63Improved
add GA-RF relation filtering on top of BHEP (ablation)HSGG · PSG zero-shot SGGzR@100: 11.63 -> 12.13Improved
add GG-CRD contrastive decoding on top of BHEP + GA-RF (ablation)HSGG · PSG zero-shot SGGzR@100: 12.13 -> 14.05Improved
add GA-RF to GG-CRD without BHEP (ablation)HSGG · PSG zero-shot SGGzR@100: 7.35 -> 7.58Improved
use HSGG vs training-free OWSGG on VG zero-shotzero-shot SGG · VG150R@50/100: +12.3% / +13.1% (R@50/100)Improved
use HSGG vs trained CAPSGG on VG zero-shotzero-shot SGG · VG150R@50/100: +1.0% / +1.8%Improved
use HSGG vs OWSGG on PSG zero-shotzero-shot SGG · PSGR@50/100: +6.7% / +7.7%Improved
use HSGG vs OvSGTR on open-world VGopen-world SGG · VG open-worldR@50/100: +3.3% / +1.9%Improved
add HSG scene graphs to InternVL3-8BInternVL3-8B · V* Benchoverall accuracy: 71.73 -> 76.43Improved
add HSG scene graphs to LLaVA-OV-1.5-8BLLaVA-OV-1.5-8B · V* Benchoverall accuracy: 76.96 -> 79.58Improved
use HSG scene graph as sole input instead of image for captioningLLaVA-OV-1.5-8B · CompreCapSunified: 59.70 -> 62.97Improved
use GG-CRD decoding instead of regular decodingLLaVA-1.5-7B · relational hallucination benchmarksaverage accuracy: +9.60% relative to regular decodingImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy