Research map / Detection and segmentation / CNN features
Single-Stage Visual Relationship Learning using Conditional Queries
Alakh Desai, Tz-Ying Wu, Subarna Tripathi, Nuno Vasconcelos
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 |
|---|---|---|---|
| Switch from shared queries (SD) to disjoint queries (DD, DDTR) | baseline SGG transformers · VG150 | mean Recall@20: 5.7 -> 9.2 | Improved |
| Disjoint queries (DDTR) instead of shared queries (DD) | baseline SGG transformers · VG150 | inference time (sec): 0.068 -> 0.220 | Hurt |
| Use conditioned queries (TraCQ) instead of DDTR | TraCQ · VG150 | mean Recall@20: 9.2 -> 12.0 | Improved |
| Use TraCQ instead of DDTR | TraCQ · VG150 | parameters (M): 82.9M -> 51.2 | Improved |
| Use entity-first variant instead of predicate-first formulation (ablation) | TraCQ · VG150 | mR@K | Hurt |
| Increase k (candidate entity boxes) from 1 to 5 (ablation) | TraCQ · VG150 | mR@20 and R@20 | Improved |
| Increase k beyond 5 (ablation) | TraCQ · VG150 | mR@20 and R@20 | No clear effect |
| Remove entity box refinement module C (use boxes from predicate decoder) (ablation) | TraCQ · VG150 PredDet | mR@K | Hurt |
| Add predicate label conditioning to entity refinement queries (ablation) | TraCQ · VG150 | mR@20: ~2 points | Improved |
| Refinement module predicts boxes and labels instead of labels only (ablation) | TraCQ · VG150 | mR@20/50/100: 10.7/11.7/12.1 -> 12.0/13.8/14.6 | Improved |
| Use TraCQ instead of DD in model size | TraCQ · VG150 | inference time vs DDTR: 1.5x faster | Improved |
Closest papers by meaning
- Hydra-SGG: Hybrid Relation Assignment for One-stage Scene Graph Generation (ICLR 2025 · similarity 0.91)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.90)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.89)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.88)
- Synthetic Visual Genome (CVPR 2025 · similarity 0.88)
- RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning (AAAI 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.
