Research map / Vision-language and multimodal / Human and embodied modeling
Attention-Aligned Transformer for Image Captioning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 perturbation-based attention alignment | Transformer image captioner · MS COCO Karpathy test, cross-entropy | CIDEr: 119.8 -> 125.0 | Helped |
| add perturbation-based attention alignment | Transformer image captioner · MS COCO Karpathy test, CIDEr optimization | CIDEr: 133.4 -> 133.9 | Helped |
| add perturbation-based attention alignment | Transformer image captioner · MS COCO online test c5 | CIDEr: 130.3 -> 132.4 | Helped |
| add perturbation-based attention alignment | Transformer image captioner · MS COCO online test c40 | CIDEr: 132.5 -> 134.7 | Helped |
| add max-pooling fusion of aligned and original attention (ablation) | Transformer image captioner · MS COCO validation | CIDEr: 120.0 -> 122.8 | Helped |
| add moving-average fusion of aligned and original attention (ablation) | Transformer image captioner · MS COCO validation | CIDEr: 120.0 -> 122.3 | Helped |
| add exponential-decay fusion of aligned and original attention (ablation) | Transformer image captioner · MS COCO validation | CIDEr: 120.0 -> 124.2 | Helped |
| add learnable-gate fusion of aligned and original attention (ablation) | Transformer image captioner · MS COCO validation | CIDEr: 120.0 -> 125.0 | Helped |
| use learnable-gate fusion rather than max-pooling fusion (ablation) | A2 Transformer · MS COCO validation | CIDEr: 122.8 -> 125.0 | Helped |
| use learnable-gate fusion rather than moving-average fusion (ablation) | A2 Transformer · MS COCO validation | CIDEr: 122.3 -> 125.0 | Helped |
| use learnable-gate fusion rather than exponential-decay fusion (ablation) | A2 Transformer · MS COCO validation | CIDEr: 124.2 -> 125.0 | Helped |
| increase the mask perturbation degree (ablation) | A2 Transformer · MS COCO validation | CIDEr | Mixed |
Closest papers by meaning
- Adaptively Aligned Image Captioning via Adaptive Attention Time (NeurIPS 2019 · similarity 0.89)
- Attention Correctness in Neural Image Captioning (AAAI 2017 · similarity 0.89)
- Text-Guided Attention Model for Image Captioning (AAAI 2017 · similarity 0.89)
- Knowing When to Look: Adaptive Attention via a Visual Sentinel for Image Captioning (CVPR 2017 · similarity 0.88)
- Visual Attention Never Fades: Selective Progressive Attention ReCalibration for Detailed Image Captioning in Multimodal Large Language Models (ICML 2025 · similarity 0.85)
- No Head Left Behind – Multi-Head Alignment Distillation for Transformers (AAAI 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.