Research map / Vision-language and multimodal / Human and embodied modeling
MemCap: Memorizing Style Knowledge for Image Captioning
Wentian Zhao, Xinxiao Wu, Xiaoxun Zhang
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 sentence decomposer for style/content split (ablation) | MemCap · SentiCap positive | CIDEr: 47.0 -> 54.4 | Improved |
| add sentence decomposer for style/content split (ablation) | MemCap · SentiCap positive | perplexity: 26.5 -> 13.0 | Improved |
| add sentence decomposer for style/content split (ablation) | MemCap · SentiCap positive | style accuracy: 63.4 -> 99.8 | Improved |
| add style memory module (ablation) | MemCap · SentiCap positive | style accuracy: 67.1 -> 99.8 | Improved |
| add style memory module (ablation) | MemCap · FlickrStyle10K romantic | perplexity: 23.1 -> 14.4 | Improved |
| add style memory module (ablation) | MemCap · SentiCap positive | CIDEr: 54.7 -> 54.4 | No clear effect |
| add self-critical training (ablation) | MemCap · SentiCap positive | CIDEr: 46.6 -> 54.4 | Improved |
| add self-critical training (ablation) | MemCap · FlickrStyle10K romantic | style accuracy: 72.4 -> 98.7 | Improved |
| add CIDEr reward (ablation) | MemCap · FlickrStyle10K romantic | CIDEr: 7.8 -> 22.4 | Improved |
| add perplexity reward (ablation) | MemCap · SentiCap positive | perplexity: 24.6 -> 13.0 | Improved |
| add style classifier reward (ablation) | MemCap · FlickrStyle10K romantic | style accuracy: 24.3 -> 98.7 | Improved |
| add style classifier reward (ablation) | MemCap · SentiCap positive | CIDEr: 56.4 -> 54.4 | Mixed |
Closest papers by meaning
- SemStyle: Learning to Generate Stylised Image Captions Using Unaligned Text (CVPR 2018 · similarity 0.90)
- ``Factual'' or ``Emotional'': Stylized Image Captioning with Adaptive Learning and Attention (ECCV 2018 · similarity 0.89)
- MSCap: Multi-Style Image Captioning With Unpaired Stylized Text (CVPR 2019 · similarity 0.89)
- StyleNet: Generating Attractive Visual Captions With Styles (CVPR 2017 · similarity 0.89)
- SCSA: A Plug-and-Play Semantic Continuous-Sparse Attention for Arbitrary Semantic Style Transfer (CVPR 2025 · similarity 0.86)
- MeaCap: Memory-Augmented Zero-shot Image Captioning (CVPR 2024 · 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.
