Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Dual Attention Networks for Few-Shot Fine-Grained Recognition
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 |
|---|---|---|---|
| use dual attention with hard-attended MIL parts and soft attention | CNN-4 few-shot fine-grained recognition · CUB | accuracy: 72.89±0.50 vs 67.97±0.44 | Helped |
| use dual attention with hard-attended MIL parts and soft attention | CNN-4 few-shot fine-grained recognition · Cars | accuracy: 70.21±0.50 vs 66.11±0.54; 85.55±0.31 vs 89.60±0.44 | Mixed |
| use dual attention with hard-attended MIL parts and soft attention | CNN-4 few-shot fine-grained recognition · Dogs | accuracy: 59.81±0.50 vs 54.84±0.53; 77.19±0.35 vs 71.90±0.68 | Helped |
| replace proposed MIL aggregation with naive global average pooling (ablation) | DUAL ATT-NET · CUB 1-shot | accuracy: 72.89 -> 71.96 | Hurt |
| replace proposed MIL aggregation with naive global average pooling (ablation) | DUAL ATT-NET · Cars 1-shot | accuracy: 70.41 -> 69.59 | Hurt |
| replace proposed MIL aggregation with naive global average pooling (ablation) | DUAL ATT-NET · Dogs 1-shot | accuracy: 59.81 -> 58.95 | Hurt |
| add global feature concatenation to the proposed MIL aggregation (ablation) | DUAL ATT-NET · CUB 1-shot | accuracy: 58.40 -> 72.89 | Helped |
| add global feature concatenation to the proposed MIL aggregation (ablation) | DUAL ATT-NET · Cars 1-shot | accuracy: 54.30 -> 70.41 | Helped |
| add global feature concatenation to the proposed MIL aggregation (ablation) | DUAL ATT-NET · Dogs 1-shot | accuracy: 50.76 -> 59.81 | Helped |
| model relationships among selected fine-grained parts rather than use identity adjacency (ablation) | DUAL ATT-NET · CUB 1-shot | accuracy: 72.22 -> 72.98 | Helped |
| model relationships among selected fine-grained parts rather than use identity adjacency (ablation) | DUAL ATT-NET · Cars 1-shot | accuracy: 69.48 -> 70.41 | Helped |
| model relationships among selected fine-grained parts rather than use identity adjacency (ablation) | DUAL ATT-NET · Dogs 1-shot | accuracy: 59.19 -> 59.81 | Helped |
Closest papers by meaning
- A Dual Attention Network with Semantic Embedding for Few-Shot Learning (AAAI 2019 · similarity 0.89)
- Looking for the Devil in the Details: Learning Trilinear Attention Sampling Network for Fine-Grained Image Recognition (CVPR 2019 · similarity 0.89)
- Dual Cross-Attention Learning for Fine-Grained Visual Categorization and Object Re-Identification (CVPR 2022 · similarity 0.88)
- A^2-Nets: Double Attention Networks (NeurIPS 2018 · similarity 0.88)
- TransFG: A Transformer Architecture for Fine-Grained Recognition (AAAI 2022 · similarity 0.88)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 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.