Research map / Video / Action recognition
A^2-Nets: Double Attention Networks
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 5 A^2-blocks to ResNet-50 | ResNet-50 · ImageNet-1k validation | Top-1 accuracy | Helped |
| Add 5 A^2-blocks to ResNet-50 | ResNet-50 · Kinetics | Video @1 | Helped |
| Transfer Kinetics-pretrained A^2-Net features | A^2-Net · UCF-101 | Video @1 | Hurt |
| Add one A^2-block at Conv2 (ablation) | ResNet-26 · Kinetics | Clip @1: 50.4 % -> 51.2 % | Helped |
| Add one A^2-block at Conv3 (ablation) | ResNet-26 · Kinetics | Clip @1: 50.4 % -> 51.9 % | Helped |
| Add one A^2-block at Conv4 (ablation) | ResNet-26 · Kinetics | Clip @1: 50.4 % -> 52.3 % | Helped |
| Place the single A^2-block at a higher stage rather than Conv2 (ablation) | ResNet-26 · Kinetics | Clip @1: 51.2 % -> 52.3 % | Helped |
| Add one A^2-block rather than one non-local block at Conv4 (ablation) | ResNet-26 · Kinetics | Clip @1: 51.7 % -> 52.3 % | Helped |
| Add 2 A^2-blocks at Conv4 instead of 1 (ablation) | ResNet-26 · Kinetics | Clip @1: 52.3 % -> 52.5 % | Helped |
| Add 4 A^2-blocks across Conv3 and Conv4 instead of 2 at Conv4 (ablation) | ResNet-26 · Kinetics | Clip @1: 52.5 % -> 53.0 % | Helped |
| Distribute 4 blocks across Conv3 and Conv4 rather than placing 4 in one stage (ablation) | A^2-Net · Kinetics | Clip @1 | Helped |
Closest papers by meaning
- Residual Attention Network for Image Classification (CVPR 2017 · similarity 0.88)
- Dual Attention Networks for Few-Shot Fine-Grained Recognition (AAAI 2022 · similarity 0.88)
- Second-Order Attention Network for Single Image Super-Resolution (CVPR 2019 · similarity 0.87)
- Attention Branch Network: Learning of Attention Mechanism for Visual Explanation (CVPR 2019 · similarity 0.86)
- Learn to Pay Attention (ICLR 2018 · similarity 0.86)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 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.