Research map / Graph learning / Knowledge and temporal graphs
HiLo: Exploiting High Low Frequency Relations for Unbiased Panoptic Scene Graph Generation
Zijian Zhou, Miaojing Shi, Holger Caesar
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 |
|---|---|---|---|
| apply HiLo framework on top of the HiLo baseline (ablation) | HiLo baseline · PSG | mR@100: 28.4 -> 33.1 | Improved |
| apply HiLo framework to PSGTR (ablation) | PSGTR · PSG | mR@100: 22.1 -> 24.5 | Improved |
| add relation augmentation to raise multi-relation pairs from 10% to 40% (ablation) | HiLo · PSG | mR@100: 32.5 -> 33.1 | Improved |
| add subject-object and relation consistency losses (ablation) | HiLo · PSG | mR@100: 32.0 -> 33.1 | Improved |
| use margin 0.5 rather than 0 in relation consistency loss (ablation) | HiLo · PSG | mR@100: 32.8 -> 33.1 | Improved |
| add relational index exchange (RIE) (ablation) | HiLo relation consistency loss · PSG | mR@100: 32.3 -> 33.1 | Improved |
| fuse H-L and L-H branch predictions at inference (ablation) | HiLo · PSG | mR@100: 30.9 -> 33.1 | Improved |
| average branch output tensors instead of inference fusion (ablation) | HiLo · PSG | mR@100: 33.1 -> 16.3 | Hurt |
| use masked relation attention on subject-object union rather than full image (ablation) | HiLo baseline · PSG | mR@100: 26.8 -> 28.4 | Improved |
| HiLo versus IETrans on VG-150 with MOTIF | MOTIF · VG-150 | R@100: 30.6 -> 30.3 | Mixed |
Later papers that compared with this method
5 later accepted papers report HiLo as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · better on 3 of 4 shared results)
- OpenPSG: Open-set Panoptic Scene Graph Generation via Large Multimodal Models (ECCV 2024 · better on 2 of 2 shared results)
- A Fair Ranking and New Model for Panoptic Scene Graph Generation (ECCV 2024 · better on 2 of 2 shared results)
- Pair Then Relation: Pair-Net for Panoptic Scene Graph Generation (TPAMI 2024 · better on 1 of 2 shared results)
- From Easy to Hard: Learning Curricular Shape-Aware Features for Robust Panoptic Scene Graph Generation (IJCV 2024 · better on 0 of 1 shared result)
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
- Pair Then Relation: Pair-Net for Panoptic Scene Graph Generation (TPAMI 2024 · similarity 0.92)
- Panoptic Scene Graph Generation (ECCV 2022 · similarity 0.92)
- A Fair Ranking and New Model for Panoptic Scene Graph Generation (ECCV 2024 · similarity 0.91)
- OpenPSG: Open-set Panoptic Scene Graph Generation via Large Multimodal Models (ECCV 2024 · similarity 0.91)
- 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 0.89)
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.
