Research map / Graph learning / Knowledge and temporal graphs
Unpaired Image Captioning via Scene Graph Alignments
Jiuxiang Gu, Shafiq Joty, Jianfei Cai, Handong Zhao, Xu Yang, Gang Wang
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 |
|---|---|---|---|
| replace average pooling with attention over features (ablation) | Graph-Enc-Dec · MSCOCO test (sentence scene graph input) | CIDEr: 129.4 -> 151.4 | Improved |
| use independent attention for each feature set instead of one shared attention (ablation) | Graph-Enc-Dec · MSCOCO test (sentence scene graph input) | CIDEr: 151.4 -> 168.8 | Improved |
| use independent attention instead of average pooling on image scene graphs (no GAN) (ablation) | Graph-Enc-Dec · MSCOCO test (unpaired, no GAN) | CIDEr: 41.4 -> 48.5 | Improved |
| independent attention changes BLEU-4 (no GAN) (ablation) | Graph-Enc-Dec · MSCOCO test (unpaired, no GAN) | BLEU-4: 20.3 -> 11.9 | Mixed |
| add CycleGAN feature alignment (GP loss, 64-d discriminator) to Graph-Enc-Dec (Att) (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 48.5 -> 69.5 | Improved |
| use gradient penalty loss instead of BCE (64-d discriminator) (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 65.5 -> 69.5 | Improved |
| use gradient penalty loss instead of MSE (64-d discriminator) (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 63.8 -> 69.5 | Improved |
| use 64-d discriminator output instead of 1-d with GP loss (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 61.1 -> 69.5 | Improved |
| use 64-d discriminator output instead of 1-d with MSE loss (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 46.7 -> 63.8 | Improved |
| use 64-d discriminator output instead of 1-d with BCE loss (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 65.2 -> 65.5 | No clear effect |
| use separate mapping per feature type instead of one shared GAN (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 60.1 -> 69.5 | Improved |
| use separate mapping per feature type instead of concatenated single GAN (ablation) | Graph-Align · MSCOCO test (unpaired) | CIDEr: 61.2 -> 69.5 | Improved |
Closest papers by meaning
- Comprehensive Image Captioning via Scene Graph Decomposition (ECCV 2020 · similarity 0.91)
- Benchmarking Large Vision-Language Models via Directed Scene Graph for Comprehensive Image Captioning (CVPR 2025 · similarity 0.87)
- Fine-Grained Captioning of Long Videos through Scene Graph Consolidation (ICML 2025 · similarity 0.87)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.86)
- Relational Distant Supervision for Image Captioning without Image-Text Pairs (AAAI 2024 · similarity 0.86)
- Composing Object Relations and Attributes for Image-Text Matching (CVPR 2024 · similarity 0.85)
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.
