Research map / Architectures and efficiency / CNN design
SCA-CNN: Spatial and Channel-Wise Attention in Convolutional 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 |
|---|---|---|---|
| Combine channel-wise and spatial attention, with channel attention applied first (ablation) | Spatial-attention captioner using ResNet-152 · Flickr8K | B@4: 20.5 -> 25.7 | Helped |
| Combine channel-wise and spatial attention, with channel attention applied first (ablation) | Spatial-attention captioner using ResNet-152 · Flickr30K | B@4: 20.5 -> 22.1 | Helped |
| Combine channel-wise and spatial attention, with channel attention applied first (ablation) | Spatial-attention captioner using ResNet-152 · MSCOCO | B@4: 28.3 -> 30.4 | Helped |
| Apply spatial attention before channel-wise attention instead of channel-wise attention first (ablation) | Combined-attention captioner using ResNet-152 · MSCOCO | B@4: 29.8 -> 30.4 | Hurt |
| Add a third spatial-attention layer (ablation) | Spatial-attention captioner using ResNet-152 · Flickr8K | B@4: 20.5 -> 23.9 | Helped |
| Add a third spatial-attention layer (ablation) | Spatial-attention captioner using ResNet-152 · Flickr30K | B@4: 20.5 -> 21.0 | Helped |
| Add a third spatial-attention layer (ablation) | Spatial-attention captioner using ResNet-152 · MSCOCO | B@4: 28.3 -> 29.6 | Helped |
| Add a third channel-spatial attention layer instead of using two layers (ablation) | C-S captioner using ResNet-152 · Flickr8K | B@4: 25.8 -> 25.3 | Hurt |
| Add a third channel-spatial attention layer instead of using two layers (ablation) | C-S captioner using ResNet-152 · Flickr30K | B@4: 22.3 -> 22.0 | Hurt |
| Add a third channel-spatial attention layer instead of using two layers (ablation) | C-S captioner using ResNet-152 · MSCOCO | B@4: 31.1 -> 30.9 | Hurt |
| Use the proposed two-layer C-S model instead of the strongest listed baseline | Image captioning · Flickr8K | B@4: 21.3 -> 25.8 | Helped |
| Use the proposed two-layer C-S model instead of the strongest listed baseline | Image captioning · Flickr30K | B@4: 23.0 -> 22.3 | Hurt |
Closest papers by meaning
- Normalized and Geometry-Aware Self-Attention Network for Image Captioning (CVPR 2020 · similarity 0.88)
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · similarity 0.88)
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.87)
- Convolutional Image Captioning (CVPR 2018 · similarity 0.87)
- Text-Guided Attention Model for Image Captioning (AAAI 2017 · similarity 0.87)
- CSTA: CNN-based Spatiotemporal Attention for Video Summarization (CVPR 2024 · similarity 0.86)
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.