Research map / Graph learning / Knowledge and temporal graphs
RepSGG: Novel Representations of Entities and Relationships for Scene Graph Generation
Hengyue Liu, Bir Bhanu
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 logit adjustment (ablation) | RepSGG · VG150 PredCls | mR@100: 27.1 -> 39.4 | Improved |
| use PGLA instead of standard LA (ablation) | RepSGG · VG150 PredCls | mR@100: 39.4 -> 40.3 | Improved |
| add margin ranking loss (ablation) | RepSGG-PGLA · VG150 PredCls | mR@100: 40.3 -> 41.2 | Improved |
| add whole-model fine-tuning (ablation) | RepSGG-PGLA · VG150 SGCls | mR@100: 21.4 -> 27.7 | Improved |
| add encoder and decoder layers (ablation) | RepSGG-PGLA · VG150 PredCls | mR@100: 39.3 -> 41.2 | Improved |
| increase number of rep-embeddings K from 1 to 10 (ablation) | RepSGG-PGLA · VG150 PredCls | mR@100: 41.0 -> 41.2 | No clear effect |
| add RCA to GCA-only decoder (ablation) | RepSGG-PGLA · VG150 PredCls | mR@100: 40.6 -> 41.4 | Improved |
| add GCA to RCA-only decoder (ablation) | RepSGG-PGLA · VG150 PredCls | mR@100: 41.4 -> 41.4 | No clear effect |
| add bias term B in PGLA (ablation) | RepSGG · VG150 PredCls | mR@100: 27.1 -> 42.0 | Improved |
| add all of W, B, D instead of B alone (ablation) | RepSGG-PGLA · VG150 PredCls | zs-mR@100: 17.7 -> 20.1 | Improved |
| add PGLA (debiasing) | RepSGG · VG150 PredCls | R@100: 65.3 -> 28.8 | Hurt |
| add PGLA (debiasing) | RepSGG · VG150 PredCls | mR@100: 24.4 -> 43.7 | Improved |
Closest papers by meaning
- Foundation Models Based Scene Graph Generation (TPAMI 2026 · similarity 0.91)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.91)
- Single-Stage Visual Relationship Learning using Conditional Queries (NeurIPS 2022 · similarity 0.90)
- Not All Relations Are Equal: Mining Informative Labels for Scene Graph Generation (CVPR 2022 · similarity 0.90)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.89)
- Hydra-SGG: Hybrid Relation Assignment for One-stage Scene Graph Generation (ICLR 2025 · 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.
