Research map / Graph learning / Knowledge and temporal graphs
Comprehensive Image Captioning via Scene Graph Decomposition
Yiwu Zhong, Liwei Wang, Jianshu Chen, Dong Yu, Yin Li
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 consensus re-ranking on top sGPN captions (ablation) | Sub-GC · MS-COCO M-RNN val | CIDEr: +4.2 CIDEr with NMS=0.75 | Improved |
| use sGPN instead of consensus re-ranking (ablation) | Sub-GC · MS-COCO M-RNN val | CIDEr: +1.1 CIDEr and +12% 1-gram with NMS=0.75 | Improved |
| lower NMS threshold from 0.75 to 0.55 (ablation) | Sub-GC-sGPN · MS-COCO M-RNN val | 1-gram diversity: 0.28 -> 0.32 | Improved |
| lower NMS threshold from 0.75 to 0.55 (ablation) | Sub-GC-consensus · MS-COCO M-RNN val | CIDEr: 107.6 -> 105.6 | Hurt |
| lower NMS threshold from 0.75 to 0.55 (ablation) | Sub-GC · MS-COCO M-RNN val | Distinct captions: 59.3% -> 70.5% | Improved |
| add top-K sampling (T=0.6, K=3) (ablation) | Sub-GC · MS-COCO M-RNN val | CIDEr vs diversity: 112.3 -> 108.7 CIDEr; 70.5% -> 96.0% distinct | Mixed |
| increase sampling randomness (T=1.0, K=3 vs T=0.6, K=3) (ablation) | Sub-GC-S · MS-COCO M-RNN val | CIDEr: 108.7 -> 100.7 | Hurt |
| decompose scene graph into sub-graphs instead of full graph | Sub-GC vs Full-GC · Flickr30k Entities grounding | F1_loc: 13.08 -> 16.53 | Improved |
| decompose scene graph into sub-graphs instead of full graph | Sub-GC vs Full-GC · Flickr30k Entities | CIDEr: 63.5 -> 61.9 | Hurt |
| decompose scene graph into sub-graphs instead of full graph | Sub-GC vs Full-GC · COCO Karpathy test | CIDEr: 114.8 -> 115.3 | No clear effect |
| add top-K sampling to Sub-GC vs Seq-CVAE comparison (Sub-GC-S) | Sub-GC-S · COCO M-RNN test (100 samples) | 1-gram diversity: 0.33 -> 0.39 | Improved |
| add region-set supervision to Sub-GC | Sub-GC · Flickr30k Entities controllable | CIDEr: 57.8 -> 73.7 | Improved |
Closest papers by meaning
- In Defense of Scene Graphs for Image Captioning (ICCV 2021 · similarity 0.93)
- Unpaired Image Captioning via Scene Graph Alignments (ICCV 2019 · similarity 0.91)
- Fine-Grained Captioning of Long Videos through Scene Graph Consolidation (ICML 2025 · similarity 0.91)
- Benchmarking Large Vision-Language Models via Directed Scene Graph for Comprehensive Image Captioning (CVPR 2025 · similarity 0.90)
- SGDiff: Scene Graph Guided Diffusion Model for Image Collaborative SegCaptioning (AAAI 2025 · similarity 0.88)
- SCENIR: Visual Semantic Clarity through Unsupervised Scene Graph Retrieval (ICML 2025 · similarity 0.88)
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.
