Research map / Graph learning / Knowledge and temporal graphs
Compositional Feature Augmentation for Unbiased Scene Graph Generation
Lin Li, Guikun Chen, Jun Xiao, Yi Yang, Chunping Wang, Long Chen
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 compositional feature augmentation (CFA) to Motifs | Motifs · VG PredCls | mR@50 / 100: 16.5 / 17.8 -> 35.7 / 38.2 | Improved |
| add CFA to Motifs | Motifs · VG PredCls | R@50 / 100: 65.6 / 67.2 -> 54.1 / 56.6 | Hurt |
| use CFA instead of NICE for debiasing | Motifs · VG PredCls | Mean: 43.6 -> 46.2 | Improved |
| add CFA to baselines on GQA | Motifs, VCTree, Transformer · GQA PredCls | mR@50 / 100: 13.9 / 14.7 -> 31.7 / 33.8 (Motifs) | Improved |
| add component prior knowledge to CFA | Motifs+CFA · VG PredCls | mR@100: 38.2 -> 43.0 | Improved |
| add intrinsic feature replacement only (ablation) | Motifs · VG PredCls | mR@K: 16.5 / 17.8 -> 19.3 / 21.2 mR; R 65.6 / 67.2 -> 64.7 / 66.6 | Mixed |
| add extrinsic mixup of foreground or background triplets (ablation) | Motifs · VG PredCls | mR@100: 17.8 -> 30.3 (R@100 67.2 -> 63.4) | Improved |
| combine all CFA components (ablation) | Motifs · VG PredCls | Mean: 41.8 -> 46.2 | Improved |
| fuse context and semantic similarity with pattern similarity in clustering (ablation) | Intrinsic-CFA · VG PredCls | Mean: 43.9 -> 46.2 | Improved |
| use 15 clusters instead of 150 (ablation) | Intrinsic-CFA · VG PredCls | mR@50 / 100: 33.2 / 35.7 -> 35.7 / 38.2 | Improved |
| set mixup theta to 0.5 rather than 0.7 or 1.0 (ablation) | Extrinsic-CFA · VG PredCls | Mean: 42.9 -> 46.2 (0.7 vs 0.5) | Improved |
Closest papers by meaning
- Multi-level Compositional Feature Augmentation for Unbiased Scene Graph Generation (IJCV 2026 · similarity 0.99)
- Fast Contextual Scene Graph Generation With Unbiased Context Augmentation (CVPR 2023 · similarity 0.91)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.90)
- RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning (AAAI 2025 · similarity 0.89)
- Adaptive Self-training Framework for Fine-grained Scene Graph Generation (ICLR 2024 · similarity 0.89)
- Hydra-SGG: Hybrid Relation Assignment for One-stage Scene Graph Generation (ICLR 2025 · 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.
