Research map / Graph learning / Knowledge and temporal graphs
A Fair Ranking and New Model for Panoptic Scene Graph Generation
Julian Lorenz, Alexander Pest, Daniel Kienzle, Katja Ludwig, Rainer Lienhart
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 |
|---|---|---|---|
| evaluate one-stage PSGG methods under fair SingleMPO instead of MultiMPO | PSGTR, PSGFormer, Pair-Net, HiLo · PSG SGGen | mR@50: HiLo 37.60 -> 18.33 | Hurt |
| replace first-stage model with MaskDINO for two-stage methods | VCTree, GPS-Net · PSG SGGen | mR@50: +7.4 for VCTree, almost doubles GPS-Net | Improved |
| add semantic token (with real masks) (ablation) | DSFormer · PSG PredCls | mR@50: 34.35 -> 36.64 (rectangular to masks plus semantic) | Improved |
| add location token (with real masks) (ablation) | DSFormer · PSG PredCls | mR@50: 36.64 -> 40.06 with both tokens; 32.48 -> 38.78 without semantic token vs without location | Improved |
| add location token with rectangular masks (ablation) | DSFormer · PSG PredCls | mR@50: 33.75 -> 33.83 | No clear effect |
| add auxiliary node loss (weight 0.2) (ablation) | DSFormer · PSG PredCls | mR@50: 35.96 -> 40.06 | Improved |
| raise node loss weight from 0.2 to 0.5 (ablation) | DSFormer · PSG PredCls | mR@50: 40.06 -> 39.90 | No clear effect |
| increase transformer embedding dimension 192 to 384 (ablation) | DSFormer · PSG PredCls | mR@50: 38.01 -> 40.06 | Improved |
| use binary instead of ratio subject-object encoding (ablation) | DSFormer · PSG | mR@50 | No clear effect |
| use segmentation model with higher PQ as first stage for DSFormer (ablation) | DSFormer · PSG SGGen | mR@50: 26.97 (Mask2Former) -> 30.67 (MaskDINO) | Improved |
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 0.93)
- Panoptic Video Scene Graph Generation (CVPR 2023 · similarity 0.92)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.90)
- Panoptic Scene Graph Generation with Semantics-Prototype Learning (AAAI 2024 · similarity 0.88)
- Scene Graph-Grounded Image Generation (AAAI 2025 · similarity 0.87)
- Towards Unbiased and Robust Spatio-Temporal Scene Graph Generation and Anticipation (CVPR 2025 · 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.
