Research map / Detection and segmentation / Open-set and few-shot detection
Graphical Contrastive Losses for Scene Graph Parsing
Ji Zhang, Kevin J. Shih, Ahmed Elgammal, Andrew Tao, Bryan Catanzaro
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 all three graphical contrastive losses to predicate cross-entropy training (ablation) | RelDN · OpenImages mini | wmAP_rel: 34.63 -> 35.54 | Improved |
| add all three graphical contrastive losses to predicate cross-entropy training (ablation) | RelDN · OpenImages mini | AP_rel plays: 36.04 -> 41.04 (+5.0) | Improved |
| add all three graphical contrastive losses (ablation) | RelDN · Visual Genome SGDET | Recall@50: 28.1 -> 28.3 | Improved |
| add all three graphical contrastive losses (ablation) | RelDN with COCO initialization · VRD relationship detection, free k | Recall@50: 26.67 -> 28.15 | Improved |
| use class-agnostic contrastive loss L1 (ablation) | RelDN · OpenImages mini | wmAP_rel: 34.63 -> 35.25 | Improved |
| use entity-class-aware contrastive loss L2 (ablation) | RelDN · OpenImages mini | wmAP_rel: 34.63 -> 35.03 | Improved |
| use predicate-class-aware contrastive loss L3 (ablation) | RelDN · OpenImages mini | wmAP_rel: 34.63 -> 35.19 | Improved |
| use all three losses together instead of L1 and L2 only (ablation) | RelDN · OpenImages mini | AP_rel plays: 38.58 -> 41.04 | Improved |
| set every contrastive margin threshold to 1.0 instead of 0.2 (ablation) | RelDN · OpenImages mini | wmAP_rel: 35.54 -> 34.17 | Hurt |
| add visual subject-predicate-object features to the semantic module (ablation) | RelDN · OpenImages mini | score_wtd: 39.32 -> 44.06 | Improved |
| add separate subject and object skip connections to the visual module (ablation) | RelDN · OpenImages mini | score_wtd: 44.06 -> 44.44 | Improved |
| add the spatial module to semantic and visual modules (ablation) | RelDN · OpenImages mini | score_wtd: 44.44 -> 44.61 | Improved |
Closest papers by meaning
- Weakly Supervised Visual Semantic Parsing (CVPR 2020 · similarity 0.90)
- Explanation-based Weakly-supervised Learning of Visual Relations with Graph Networks (ECCV 2020 · similarity 0.90)
- Joint Modeling of Visual Objects and Relations for Scene Graph Generation (NeurIPS 2021 · similarity 0.90)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.89)
- RelTR: Relation Transformer for Scene Graph Generation (TPAMI 2023 · similarity 0.89)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 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.
