Research map / Detection and segmentation / CNN features
Looking for the Devil in the Details: Learning Trilinear Attention Sampling Network for Fine-Grained Image Recognition
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 |
|---|---|---|---|
| Use trilinear attention instead of raw feature maps (ablation) | TASN part-net · CUB-200-2011 | accuracy: 83.5 -> 84.9 | Helped |
| Add both spatial and relationship normalization to trilinear attention (ablation) | TASN part-net · CUB-200-2011 | accuracy: 84.9 -> 85.3 | Helped |
| Use the N(X)N(X)ᵀX spatial-normalization variant (ablation) | TASN part-net · CUB-200-2011 | accuracy: 85.3 -> 84.3 | Hurt |
| Use the normalized trilinear attention selected by the authors (ablation) | TASN · CUB-200-2011 | accuracy: 85.0 -> 85.3 | Helped |
| Use the proposed non-uniform attention sampler instead of uniform sampling (ablation) | TASN · CUB-200-2011 | accuracy: 85.8 -> 87.0 | Helped |
| Use the proposed sampler instead of the SSN sampler (ablation) | TASN · CUB-200-2011 | accuracy: 85.3 -> 87.0 | Helped |
| Add part-net feature distillation to master-net (ablation) | TASN · CUB-200-2011, resolution 224 | accuracy: 85.5 -> 87.0 | Helped |
| Increase input resolution from 336 to 392 without the distiller (ablation) | Master-net · CUB-200-2011 | accuracy: 87.0 -> 86.8 | Hurt |
| Select attention maps by response ranking rather than randomly (ablation) | TASN · CUB-200-2011 | accuracy: 87.0 -> 86.8 | Hurt |
| Distill two parts per iteration instead of one (ablation) | TASN · CUB-200-2011 | accuracy | No clear effect |
| Use TASN rather than SSN | Fine-grained classifier · iNaturalist-2017 total | accuracy: 65.2 -> 68.2 | Helped |
| Use TASN rather than ResNet baseline | Fine-grained classifier · iNaturalist-2017 total | accuracy: 59.6 -> 68.2 | Helped |
Closest papers by meaning
- Look Closer to See Better: Recurrent Attention Convolutional Neural Network for Fine-Grained Image Recognition (CVPR 2017 · similarity 0.90)
- Dual Attention Networks for Few-Shot Fine-Grained Recognition (AAAI 2022 · similarity 0.89)
- Dual Cross-Attention Learning for Fine-Grained Visual Categorization and Object Re-Identification (CVPR 2022 · similarity 0.88)
- TransFG: A Transformer Architecture for Fine-Grained Recognition (AAAI 2022 · similarity 0.87)
- Context-aware Attentional Pooling (CAP) for Fine-grained Visual Classification (AAAI 2021 · similarity 0.87)
- Exploring Self-Attention for Image Recognition (CVPR 2020 · similarity 0.87)
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.