Research map / Image generation / Text-to-image diffusion
Learning Combinatorial Prompts for Universal Controllable Image Captioning
Zhen Wang, Jun Xiao, Yueting Zhuang, Fei Gao, Jian Shao, Long Chen
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 mask attention (vs plain self-attention) (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 135.6 -> 227.5 | Improved |
| add mask on content variables (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 186.3 -> 227.5 | Improved |
| use mask attention instead of segment embedding (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 220.5 -> 227.5 | Improved |
| add segment embedding on top of mask attention (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 227.5 -> 225.2 | Hurt |
| use bottom-up region features instead of CLIP ViT features (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 193.9 -> 227.5 | Improved |
| use bottom-up features instead of CLIP features (ablation) | ComPro · COCO Entities, structure control | CIDEr-D: 132.7 -> 134.1 | No clear effect |
| increase structure prompt length from 1 to 40 (ablation) | ComPro · COCO Entities, structure control | CIDEr-D: 123.8 -> 134.1 | Improved |
| increase structure prompt length from 40 to 80 (ablation) | ComPro · COCO Entities, structure control | CIDEr-D: 134.1 -> 133.9 | No clear effect |
| increase content prompt length from 1 to 10 (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 217.2 -> 227.5 | Improved |
| increase content prompt length from 10 to 20 (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 227.5 -> 223.9 | Hurt |
| add structure control to content control (combined, large) (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 232.2 -> 235.7 | Improved |
| add tense control to combined content and structure control (ablation) | ComPro · COCO Entities, content control | CIDEr-D: 235.7 -> 269.4 | Improved |
Closest papers by meaning
- Human-Like Controllable Image Captioning With Verb-Specific Semantic Roles (CVPR 2021 · similarity 0.90)
- Controllable Image Captioning via Prompting (AAAI 2023 · similarity 0.90)
- COMMA: Co-articulated Multi-Modal Learning (AAAI 2024 · similarity 0.86)
- FlexCap: Describe Anything in Images in Controllable Detail (NeurIPS 2024 · similarity 0.86)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.85)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · 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.
