Research map / Vision-language and multimodal / Human and embodied modeling
Attention on Attention for Image Captioning
Lun Huang, Wenmin Wang, Jie Chen, Xiao-Yong Wei
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 refining module without AoA to encoder (ablation) | Base · MS-COCO Karpathy (XE) | CIDEr-D: 109.5 -> 112.5 | Improved |
| add AoA to the encoder refining module (ablation) | Base + refining module · MS-COCO Karpathy (XE) | CIDEr-D: 112.5 -> 114.5 | Improved |
| model decoder context vector with LSTM (ablation) | Base · MS-COCO Karpathy (XE) | CIDEr-D: 109.5 -> 113.5 | Improved |
| model decoder context vector with AoA instead of linear transform (ablation) | Base · MS-COCO Karpathy (XE) | CIDEr-D: 109.5 -> 113.8 | Improved |
| replace LSTM context with AoA (ablation) | Decoder · MS-COCO Karpathy (XE) | CIDEr-D: 113.5 -> 113.8 | Improved |
| stack AoA on top of LSTM (ablation) | Decoder · MS-COCO Karpathy (XE) | CIDEr-D: unstable training process | Hurt |
| use multi-head attention in decoder instead of single attention (ablation) | Base decoder · MS-COCO Karpathy (XE) | CIDEr-D: 109.5 -> 109.6 | No clear effect |
| add AoA to multi-head attention decoder (ablation) | Decoder with MH-Att · MS-COCO Karpathy (XE) | CIDEr-D: 109.6 -> 114.3 | Improved |
| add AoA to multi-head attention decoder with LSTM (ablation) | Decoder · MS-COCO Karpathy (XE) | CIDEr-D: 113.8 -> 114.3 | Improved |
| combine AoA in both encoder and decoder (full AoANet) (ablation) | Base · MS-COCO Karpathy (XE) | CIDEr-D: 109.5 -> 119.8 | Improved |
| use decoder with AoA in video captioning | Base decoder · MSR-VTT | CIDEr-D: 38.83 -> 42.44 | Improved |
| use decoder with AoA in video captioning | Base decoder · MSR-VTT | BLEU-4: 33.53 -> 37.22 | Improved |
Closest papers by meaning
- Adaptively Aligned Image Captioning via Adaptive Attention Time (NeurIPS 2019 · similarity 0.88)
- Attention-Aligned Transformer for Image Captioning (AAAI 2022 · similarity 0.88)
- Deliberate Attention Networks for Image Captioning (AAAI 2019 · similarity 0.88)
- Areas of Attention for Image Captioning (ICCV 2017 · similarity 0.87)
- Text-Guided Attention Model for Image Captioning (AAAI 2017 · similarity 0.87)
- Visual Attention Never Fades: Selective Progressive Attention ReCalibration for Detailed Image Captioning in Multimodal Large Language Models (ICML 2025 · similarity 0.84)
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.
