Research map / Vision-language and multimodal / Human and embodied modeling
X-Linear Attention Networks for Image Captioning
Check 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 |
|---|---|---|---|
| Replace conventional decoder attention with X-Linear attention (ablation) | X-LAN with Faster R-CNN image features · COCO Karpathy test split | CIDEr: 114.1 -> 117.0 | Helped |
| Add 1 X-Linear attention block to the image encoder (ablation) | X-LAN with decoder X-Linear attention · COCO Karpathy test split | CIDEr: 117.0 -> 118.2 | Helped |
| Increase encoder depth from 1 to 2 X-Linear attention blocks (ablation) | X-LAN · COCO Karpathy test split | CIDEr: 118.2 -> 119.4 | Helped |
| Increase encoder depth from 2 to 3 X-Linear attention blocks (ablation) | X-LAN · COCO Karpathy test split | CIDEr: 119.4 -> 120.0 | Helped |
| Increase encoder depth from 3 to 4 X-Linear attention blocks (ablation) | X-LAN · COCO Karpathy test split | CIDEr: 120.0 -> 120.4 | No clear effect |
| Add ELU to X-Linear attention in the encoder and decoder (ablation) | X-LAN with 4 encoder blocks · COCO Karpathy test split | CIDEr: 120.4 -> 122.0 | Helped |
| Add X-Linear attention blocks to a Transformer encoder-decoder | Transformer · COCO Karpathy test split | CIDEr: 128.3 -> 132.8 | Helped |
Closest papers by meaning
- Bilinear Attention Networks (NeurIPS 2018 · similarity 0.87)
- Text-Guided Attention Model for Image Captioning (AAAI 2017 · similarity 0.87)
- Hierarchical Attention Network for Image Captioning (AAAI 2019 · similarity 0.86)
- XAttention: Block Sparse Attention with Antidiagonal Scoring (ICML 2025 · similarity 0.86)
- SEA: Sparse Linear Attention with Estimated Attention Mask (ICLR 2024 · similarity 0.85)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 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.