Research map / Vision-language and multimodal / Human and embodied modeling
Show, Recall, and Tell: Image Captioning with Recall Mechanism
Li Wang, Zechen Bai, Yonghua Zhang, Hongtao Lu
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 semantic guide to Up-Down (ablation) | Up-Down · MSCOCO Karpathy test split, cross-entropy | CIDEr: 113.5 -> 115.3 | Improved |
| Add recalled-word slot to the semantic-guide model (ablation) | Up-Down with SG · MSCOCO Karpathy test split, cross-entropy | CIDEr: 115.3 -> 116.9 | Improved |
| Add recalled-word reward (ablation) | Up-Down with SG+RWS · MSCOCO Karpathy test split, cross-entropy | CIDEr: 116.9 -> 116.9 | No clear effect |
| Add semantic guide to Up-Down (ablation) | Up-Down · MSCOCO Karpathy test split, CIDEr optimization | CIDEr: 120.1 -> 127.3 | Improved |
| Add recalled-word slot to the semantic-guide model (ablation) | Up-Down with SG · MSCOCO Karpathy test split, CIDEr optimization | CIDEr: 127.3 -> 128.3 | Improved |
| Add recalled-word reward (ablation) | Up-Down with SG+RWS · MSCOCO Karpathy test split, CIDEr optimization | CIDEr: 128.3 -> 129.1 | Improved |
| Retrieve 5 captions rather than 1 per image (ablation) | SG+RWS captioning model · MSCOCO Karpathy validation set, cross-entropy | CIDEr: 115.8 -> 116.7 | Improved |
| Retrieve 15 rather than 5 captions per image (ablation) | SG+RWS captioning model · MSCOCO Karpathy validation set, cross-entropy | CIDEr: 116.7 -> 115.6 | Hurt |
| Set reward trade-off lambda to 0.5 rather than 0.1 (ablation) | SG+RWS+WR · MSCOCO Karpathy validation set, CIDEr optimization | CIDEr: 127.5 -> 128.9 | Improved |
| Set reward trade-off lambda to 0.7 rather than 0.5 (ablation) | SG+RWS+WR · MSCOCO Karpathy validation set, CIDEr optimization | CIDEr: 128.9 -> 127.8 | Hurt |
| Add the recall mechanism | att2in · MSCOCO Karpathy test split, CIDEr optimization | CIDEr: 119.1 -> 122.0 | Improved |
| Add the recall mechanism | att2all · MSCOCO Karpathy test split, CIDEr optimization | CIDEr: 121.7 -> 125.0 | Improved |
Closest papers by meaning
- Guiding Image Captioning Models Toward More Specific Captions (ICCV 2023 · similarity 0.84)
- Captured by Captions: On Memorization and its Mitigation in CLIP Models (ICLR 2025 · similarity 0.83)
- Noise-Aware Image Captioning with Progressively Exploring Mismatched Words (AAAI 2024 · similarity 0.82)
- EVCap: Retrieval-Augmented Image Captioning with External Visual-Name Memory for Open-World Comprehension (CVPR 2024 · similarity 0.82)
- CaptionQA: Is Your Caption as Useful as the Image Itself? (CVPR 2026 · similarity 0.82)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.82)
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.
