Research map / Multi-agent systems and games / Cooperative multi-agent RL
Counterfactual Critic Multi-Agent Training for Scene Graph Generation
Long Chen, Hanwang Zhang, Jun Xiao, Xiangnan He, Shiliang Pu, Shih-Fu Chang
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 |
|---|---|---|---|
| train with Recall@20 graph-level reward instead of XE only (ablation) | XE pretrained model · Visual Genome SGCls | R@20: 34.08 -> 35.93 | Improved |
| use SPICE instead of Recall@K as reward (ablation) | CMAT · Visual Genome SGCls | R@20: 35.93 -> 35.27 | Hurt |
| add counterfactual baseline instead of moving average (ablation) | CMAT · Visual Genome SGCls | R@20: 34.76 -> 35.93 | Improved |
| add counterfactual baseline instead of self-critical (ablation) | CMAT · Visual Genome SGCls | R@20: 34.68 -> 35.93 | Improved |
| add counterfactual baseline instead of moving average (ablation) | CMAT · Visual Genome SGDet | R@20: 16.07 -> 16.53 | Improved |
| add counterfactual baseline instead of self-critical (ablation) | CMAT · Visual Genome SGDet | R@20: 16.37 -> 16.53 | Improved |
| moving-average baseline vs XE only (ablation) | XE pretrained model · Visual Genome SGDet | R@20: 16.23 -> 16.07 | Hurt |
| increase communication steps from 2 to 5 (ablation) | CMAT · Visual Genome SGCls | R@20: 35.09 -> 35.93 | Improved |
| increase communication steps from 2 to 5 (ablation) | CMAT · Visual Genome SGDet | R@20: 16.35 -> 16.53 | Improved |
| CMAT training vs strongest baseline KER | CMAT · Visual Genome mean (graph constraint) | Mean recall: 44.1 -> 45.4 | Improved |
| CMAT training vs strongest baseline KER | CMAT · Visual Genome mean (no graph constraint) | Mean recall: 55.4 -> 57.0 | Improved |
Closest papers by meaning
- Scene Graph Disentanglement and Composition for Generalizable Complex Image Generation (NeurIPS 2024 · similarity 0.85)
- R3CD: Scene Graph to Image Generation with Relation-Aware Compositional Contrastive Control Diffusion (AAAI 2024 · similarity 0.85)
- Structure Your Data: Towards Semantic Graph Counterfactuals (ICML 2024 · similarity 0.85)
- Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models (ICML 2026 · similarity 0.84)
- Robust Stochastic Graph Generator for Counterfactual Explanations (AAAI 2024 · similarity 0.84)
- ESCA: Contextualizing Embodied Agents via Scene-Graph Generation (NeurIPS 2025 · similarity 0.84)
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.
