Research map / Graph learning / Knowledge and temporal graphs
Relationformer: A Unified Framework for Image-to-Graph Generation
Suprosanna Shit, Rajat Koner, Bastian Wittmann, Johannes Paetzold, Ivan Ezhov, Hongwei Li, Jiazhen Pan, Sahand Sharifzadeh, Georgios Kaissis, Volker Tresp, Bjoern Menze
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 the [rln]-token (linear relation head) (ablation) | Relationformer · Visual Genome SGDet | Recall@50: 22.0 -> 25.4 | Improved |
| add the [rln]-token (MLP relation head) (ablation) | Relationformer · Visual Genome SGDet | Recall@50: 26.4 -> 28.4 | Improved |
| add the [rln]-token (linear head) (ablation) | Relationformer · Visual Genome | AP@50: 24.1 -> 25.3 | Improved |
| use an MLP instead of a linear relation head (with rln-token) (ablation) | Relationformer · Visual Genome SGDet | Recall@50: 25.4 -> 28.4 | Improved |
| add the [rln]-token (ablation) | Relationformer · Synthetic vessel (3D) | Node mAP: 75.5 -> 78.5 | Improved |
| add the [rln]-token (ablation) | Relationformer · Synthetic vessel (3D) | Edge mAP: 76.3 -> 78.1 | Improved |
| increase training data from 25% to 100% (ablation) | Relationformer · Synthetic vessel (3D) | Node mAP: 17.0 -> 78.5 | Improved |
| increase training data from 50% to 100% (ablation) | Relationformer · Synthetic vessel (3D) | Node mAP: 39.2 -> 78.5 | Improved |
| increase [obj]-tokens from 75 to 200 (ablation) | Relationformer · Visual Genome SGDet | Recall@50: 26.1 -> 28.4 | Improved |
| increase [obj]-tokens from 200 to 300 (ablation) | Relationformer · Visual Genome SGDet | Recall@50: 28.4 -> 27.9 | No clear effect |
| increase transformer layers from 4 to 6 (ablation) | Relationformer · Visual Genome SGDet | Recall@50: 26.5 -> 28.4 | Improved |
| add relation prediction on top of deformable DETR (MLP head, rln-token) (ablation) | Relationformer · Visual Genome | AP@50: 26.4 -> 26.3 | No clear effect |
Later papers that compared with this method
2 later accepted papers report Relationformer as a baseline in their results tables; 2 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 8 of 9 shared results)
- Any2Graph: Deep End-To-End Supervised Graph Prediction With An Optimal Transport Loss (NeurIPS 2024 · better on 2 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Sat2Graph — Sat2Graph: Road Graph Extraction through Graph-Tensor Encoding (ECCV 2020) · better on 1 of 1 shared result
- GraphRNN — GraphRNN: Generating Realistic Graphs with Deep Auto-regressive Models (ICML 2018) · better on 1 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
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.91)
- RelTR: Relation Transformer for Scene Graph Generation (TPAMI 2023 · similarity 0.91)
- Joint Modeling of Visual Objects and Relations for Scene Graph Generation (NeurIPS 2021 · similarity 0.90)
- Attentive Relational Networks for Mapping Images to Scene Graphs (CVPR 2019 · similarity 0.90)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.90)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.88)
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.
