Research map / Vision-language and multimodal / Human and embodied modeling
A Neural Compositional Paradigm for Image Captioning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 predicted noun-phrases with ground-truth noun-phrases (ablation) | CompCap · MS-COCO offline | SPICE: 19.9 -> 36.8 | Helped |
| Replace predicted noun-phrases with ground-truth noun-phrases (ablation) | CompCap · MS-COCO offline | CIDEr: 86.2 -> 122.2 | Helped |
| Replace predicted noun-phrases with ground-truth noun-phrases (ablation) | CompCap · Flickr30k | SPICE: 14.9 -> 31.9 | Helped |
| Replace predicted noun-phrases with ground-truth noun-phrases (ablation) | CompCap · Flickr30k | CIDEr: 42.0 -> 89.7 | Helped |
| Use a ground-truth composing order after providing ground-truth noun-phrases (ablation) | CompCap + GT_np · MS-COCO offline | CIDEr: 122.2 -> 182.6 | Helped |
| Use a ground-truth composing order after providing ground-truth noun-phrases (ablation) | CompCap + GT_np · MS-COCO offline | SPICE: 36.8 -> 33.8 | Hurt |
| Use a ground-truth composing order after providing ground-truth noun-phrases (ablation) | CompCap + GT_np · Flickr30k | CIDEr: 89.7 -> 132.8 | Helped |
| Use a ground-truth composing order after providing ground-truth noun-phrases (ablation) | CompCap + GT_np · Flickr30k | SPICE: 31.9 -> 29.8 | Hurt |
| Give the left and right phrase encoders independent rather than shared parameters (ablation) | CompCap C-Module · MS-COCO | SPICE: 18.9 -> 19.9 | Helped |
| Increase the maximum number of noun-phrases (ablation) | CompCap · MS-COCO | SPICE | Helped |
| Increase the maximum number of noun-phrases (ablation) | CompCap · MS-COCO | CIDEr | Hurt |
| Train the composition modules with less training data while keeping noun-phrase classifiers trained on full data (ablation) | CompCap · MS-COCO and Flickr30k | SPICE | Helped |
Closest papers by meaning
- FINECAPTION: Compositional Image Captioning Focusing on Wherever You Want at Any Granularity (CVPR 2025 · similarity 0.88)
- Learning Visual Composition through Improved Semantic Guidance (CVPR 2025 · similarity 0.87)
- Panoptic Captioning: An Equivalence Bridge for Image and Text (NeurIPS 2025 · similarity 0.87)
- Enhancing Compositional Reasoning in CLIP via Reconstruction and Alignment of Text Descriptions (NeurIPS 2025 · similarity 0.87)
- Natural Language Inference Improves Compositionality in Vision-Language Models (ICLR 2025 · similarity 0.87)
- Enhancing Vision-Language Compositional Understanding with Multimodal Synthetic Data (CVPR 2025 · similarity 0.87)
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.