Research map / Graph learning / Knowledge and temporal graphs
OpenPSG: Open-set Panoptic Scene Graph Generation via Large Multimodal Models
Zijian Zhou, Zheng Zhu, Holger Caesar, Miaojing Shi
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 |
|---|---|---|---|
| extract pair features with RelQ-Former attention instead of mask pooling (ablation) | OpenPSG · PSG PredCls (closed-set) | R@100: 74.1 -> 79.3 | Improved |
| add selector that filters subject-object pairs (theta=0.35) (ablation) | OpenPSG · PSG PredCls (closed-set) | R@100 and runtime: 79.6 -> 79.3 R@100, 20x faster | No clear effect |
| add relation existence estimation loss (ablation) | OpenPSG · PSG PredCls (closed-set) | R@100: 78.5 -> 79.3 | Improved |
| use OpenSeeD instead of Mask2Former as segmenter (ablation) | OpenPSG · PSG SGDet (closed-set) | R@100: 48.4 -> 51.9 | Improved |
| use OpenSeeD instead of Mask2Former as segmenter (ablation) | OpenPSG · PSG PredCls (closed-set) | R@100: 78.9 -> 79.3 | No clear effect |
| use judgement instruction instead of generation instruction (ablation) | OpenPSG · PSG PredCls (open-set, 7:3) | R@100: 57.1 -> 61.5 | Improved |
| use judgement instruction instead of generation instruction with 70% novel relations (ablation) | OpenPSG · PSG PredCls (open-set, 3:7) | R@100: 18.5 -> 35.1 | Improved |
| increase RelQ-Former layers from 1 to 2 (ablation) | OpenPSG · PSG PredCls (closed-set) | R@100: 68.6 -> 79.3 | Improved |
| increase RelQ-Former layers from 2 to 4 (ablation) | OpenPSG · PSG PredCls (closed-set) | R@100: 79.3 -> 79.8 | No clear effect |
| increase RelQ-Former layers from 4 to 6 (ablation) | OpenPSG · PSG PredCls (closed-set) | R@100: 79.8 -> 78.3 | Hurt |
| fuse generation and judgement instruction outputs (ablation) | OpenPSG · PSG SGDet (open-set) | R@100: 36.7 -> 37.2 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HiLo — HiLo: Exploiting High Low Frequency Relations for Unbiased Panoptic Scene Graph Generation (ICCV 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Panoptic Scene Graph Generation (ECCV 2022 · similarity 0.92)
- Pair Then Relation: Pair-Net for Panoptic Scene Graph Generation (TPAMI 2024 · similarity 0.92)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.89)
- Panoptic Scene Graph Generation with Semantics-Prototype Learning (AAAI 2024 · similarity 0.88)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.88)
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · similarity 0.87)
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.
