Research map / Graph learning / Knowledge and temporal graphs
Grounding Consistency: Distilling Spatial Common Sense for Precise Visual Relationship Detection
Markos Diomataris, Nikolaos Gkanatsios, Vassilis Pitsikalis, Petros Maragos
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 GCD to six scene graph generators | six SGG baselines · VRD / VG200 | mP+, f-mP+ (relative): up to 35% relative on VRD and 70% on VG200 for UVTransE | Improved |
| add GCD to six scene graph generators | six SGG baselines · VRD / VG200 | HarMean (relative): up to 22% relative on VRD and 49% on VG200 | Improved |
| add GCD to ATR-Net | ATR-Net · VRD PredDet | HarMean: 46.38 -> 52.56 | Improved |
| add GCD to RelDN | RelDN · VG200 PredCls | R@50: 57.83 -> 55.49 | Hurt |
| add GCD to RelDN | RelDN · VG200 PredCls | HarMean: 25.89 -> 34.43 | Improved |
| add the Grounder to distillation (GCD vs GCD-G) (ablation) | GCD · 6 baselines avg | HarMean (relative): +4.03 -> +14.61 | Improved |
| add distillation to grounding loss (GCD vs GCD-D) (ablation) | GCD · 6 baselines avg | R@50 (relative): -2.41 -> -0.33 | Improved |
| add distillation to grounding loss (GCD vs GCD-D) (ablation) | GCD · 6 baselines avg | HarMean (relative): +10.22 -> +14.61 | Improved |
| add Grounder as source of spatial knowledge instead of spatial baseline (SpatDistill) (ablation) | GCD · 6 baselines avg | HarMean (relative): +8.82 -> +14.66 | Improved |
| add grounding consistency instead of graphical contrastive losses (GraphL) (ablation) | GCD · 6 baselines avg | HarMean (relative): +13.35 -> +14.66 | Improved |
| use oracle NCE negatives instead of Grounder (ablation) | GCD · 6 baselines avg | HarMean (relative): +14.66 -> +24.03 | Improved |
Closest papers by meaning
- Bridging Visual and Textual Semantics: Towards Consistency for Unbiased Scene Graph Generation (TPAMI 2024 · similarity 0.90)
- Scene Graph-Grounded Image Generation (AAAI 2025 · similarity 0.88)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.87)
- Synthetic Visual Genome (CVPR 2025 · similarity 0.87)
- Adaptive Visual Scene Understanding: Incremental Scene Graph Generation (NeurIPS 2024 · similarity 0.87)
- Panoptic Scene Graph Generation with Semantics-Prototype Learning (AAAI 2024 · 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.
