Research map / Graph learning / Knowledge and temporal graphs
RelTR: Relation Transformer for Scene Graph Generation
Yuren Cong, Michael Ying Yang, Bodo Rosenhahn
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 feature encoder layers (0 -> 6) (ablation) | RelTR · Visual Genome SGDET | R@50: 23.3 -> 27.5 | Improved |
| increase triplet decoder layers 3 -> 6 (ablation) | RelTR · Visual Genome SGDET | R@50: 25.9 -> 27.5 | Improved |
| increase triplet decoder layers 6 -> 9 (ablation) | RelTR · Visual Genome SGDET | R@50: 27.5 -> 27.1 | Hurt |
| increase encoder layers 6 -> 9 (ablation) | RelTR · Visual Genome SGDET | R@50: 27.5 -> 27.7 | No clear effect |
| add triplet decoder on top of entity decoder (ablation) | RelTR · Visual Genome SGDET | R@50: 18.3 -> 27.5 | Improved |
| add CSA to DVA-only decoder (ablation) | RelTR · Visual Genome SGDET | R@50: 20.9 -> 26.6 | Improved |
| add DEA to CSA+DVA decoder (ablation) | RelTR · Visual Genome SGDET | R@50: 26.6 -> 27.2 | Improved |
| add mask head for attention heat maps (ablation) | RelTR · Visual Genome SGDET | R@50: 27.2 -> 27.5 | No clear effect |
| use IoU-based assignment (threshold T<1) in set prediction loss (ablation) | RelTR · Visual Genome SGDET | R@50 | Improved |
| vary number of triplet queries N_t (ablation) | RelTR · Visual Genome SGDET | R@50 | Mixed |
| add bi-level resampling | RelTR · Visual Genome SGDET | R@50 / mR@50: R@50 27.5 -> 24.1; mR@50 10.8 -> 13.9 | Mixed |
| add logit adjustment | RelTR · Visual Genome SGDET | R@50 / mR@50: mR@50 10.8 -> 14.2; R@50 27.5 -> 25.9 | Mixed |
Later papers that compared with this method
3 later accepted papers report RelTR as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- End-to-End Hyper-Relational Information Extraction for Engineering Diagrams via Dynamically Tokenized Relation Transformer (CVPR 2026 · better on 2 of 2 shared results)
- Robo-SGG: Exploiting Layout-Oriented Normalization and Restitution Can Improve Robust Scene Graph Generation (CVPR 2026 · better on 1 of 1 shared result)
- Multi-Granularity Sparse Relationship Matrix Prediction Network for End-to-End Scene Graph Generation (ECCV 2024 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- SGTR — SGTR: End-to-End Scene Graph Generation With Transformer (CVPR 2022) · better on 1 of 2 shared results
- GPS-Net — GPS-Net: Graph Property Sensing Network for Scene Graph Generation (CVPR 2020) · 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
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.92)
- Vision Relation Transformer for Unbiased Scene Graph Generation (ICCV 2023 · similarity 0.91)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.91)
- Relationformer: A Unified Framework for Image-to-Graph Generation (ECCV 2022 · similarity 0.91)
- IS-GGT: Iterative Scene Graph Generation With Generative Transformers (CVPR 2023 · similarity 0.90)
- Scene Graph Generation by Iterative Message Passing (CVPR 2017 · similarity 0.90)
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.
