Research map / Vision-language and multimodal / Human and embodied modeling
Towards Accurate Text-Based Image Captioning With Content Diversity Exploration
Guanghui Xu, Shuaicheng Niu, Mingkui Tan, Yucheng Luo, Qing Du, Qi Wu
Check 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 |
|---|---|---|---|
| Use Anchor-Captioner’s anchor selection, ACGs, and text-specific refinement | Text-based image captioning · TextCaps validation set | CIDEr: 89.6 to 95.5 | Improved |
| Generate multiple ACG-guided captions rather than beam-search captions | Text-based image captioning · TextCaps validation set | Cover Ratio: 27.3 to 37.8 | Improved |
| Use RNN sequence projection rather than independent FC projection for ACG construction (ablation) | Anchor-Captioner · TextCaps validation set | CIDEr: 90.3 to 95.5 | Improved |
| Use RNN sequence projection rather than Transformer multiple projection for ACG construction (ablation) | Anchor-Captioner · TextCaps validation set | CIDEr: 90.7 to 95.5 | Improved |
| Use learned AnPM graph construction rather than rule-based anchor and grouping heuristics (ablation) | Anchor-Captioner · TextCaps validation set | CIDEr: 78.0 to 95.5 | Improved |
| Use ground-truth ACGs instead of predicted ACGs (ablation) | Anchor-Captioner · TextCaps validation set | CIDEr: 95.5 to 104.9 | Improved |
| Add ACG guidance to M4C-Captioner (ablation) | M4C-Captioner · TextCaps validation set | CIDEr: 89.6 to 93.8 | Improved |
| Use ground-truth ACGs rather than predicted ACGs with M4C-Captioner (ablation) | M4C-Captioner · TextCaps validation set | CIDEr: 93.8 to 99.6 | Improved |
| Train jointly on COCO and TextCaps rather than only COCO (ablation) | Anchor-Captioner · COCO validation | CIDEr: 110.3 to 96.3 | Hurt |
| Train jointly on COCO and TextCaps rather than only TextCaps (ablation) | Anchor-Captioner · TextCaps validation | CIDEr: 95.5 to 90.0 | Hurt |
Closest papers by meaning
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.88)
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.88)
- Panoptic Captioning: An Equivalence Bridge for Image and Text (NeurIPS 2025 · similarity 0.88)
- ViPCap: Retrieval Text-Based Visual Prompts for Lightweight Image Captioning (AAAI 2025 · similarity 0.87)
- Patch Matters: Training-free Fine-grained Image Caption Enhancement via Local Perception (CVPR 2025 · similarity 0.87)
- Image Captioning with Multi-Context Synthetic Data (AAAI 2024 · 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.
