Research map / Detection and segmentation / CNN features
Structured Sparse R-CNN for Direct Scene Graph Generation
Yao Teng, Limin Wang
Check 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 relation feature vectors and pairwise fusion (ablation) | Structured Sparse R-CNN · Visual Genome SGDet | R@100: 35.34 -> 36.58 | Improved |
| Add E2R fusion to relation features and pairwise fusion (ablation) | Structured Sparse R-CNN · Visual Genome SGDet | mR@100: 9.83 -> 10.04 | Improved |
| Train with pseudo-label assignment instead of full background assignment (ablation) | Structured Sparse R-CNN · Visual Genome SGDet | R@20: 24.62 -> 25.82 | Improved |
| Use pseudo-labels instead of no background labels (ablation) | Structured Sparse R-CNN · Visual Genome SGDet | R@100: 38.29 -> 37.42 | Hurt |
| Use category-specific adaptive gamma rather than fixed gamma (ablation) | Structured Sparse R-CNN · Visual Genome SGDet | mR@100: 9.76 -> 10.04 | Improved |
| Increase triplet queries from 300 to 800 (ablation) | Structured Sparse R-CNN · Visual Genome SGDet | R@100: 36.9 -> 38.4 | Improved |
| Apply logit adjustment | Structured Sparse R-CNN · Open Images V4 | mR@50: 72.62 -> 79.23 | Improved |
| Apply logit adjustment | Structured Sparse R-CNN · Open Images V6 | score_wtd: 49.38 -> 48.89 | Hurt |
| Use the direct structured sparse model instead of BGNN | Visual Genome SGDet · Visual Genome SGDet | R@100: 35.8 -> 36.9; mR@100: 12.6 -> 10.0 | Mixed |
Closest papers by meaning
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.90)
- Generating Triples With Adversarial Networks for Scene Graph Construction (AAAI 2018 · similarity 0.90)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.89)
- EGTR: Extracting Graph from Transformer 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)
- Adaptive Visual Scene Understanding: Incremental Scene Graph Generation (NeurIPS 2024 · 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.
