Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Attentive Pairwise Interaction for Fine-Grained Classification
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 attentive pairwise interaction to ResNet-101 | ResNet-101 fine-grained classifier · CUB-200-2011 | accuracy: 85.4 -> 88.6 | Helped |
| replace individual-only gate learning with a mutual vector from both images (ablation) | API-Net · CUB-200-2011 | accuracy: 87.9 -> 88.6 | Helped |
| generate mutual vector with MLP rather than compact bilinear pooling (ablation) | API-Net · CUB-200-2011 | accuracy: 88.2 -> 88.6 | Helped |
| use image-specific paired gates instead of one shared gate (ablation) | API-Net · CUB-200-2011 | accuracy: 87.7 -> 88.6 | Helped |
| add score-ranking regularization (ablation) | API-Net · CUB-200-2011 | accuracy: 88.1 -> 88.6 | Helped |
| construct pairs from class and feature similarity rather than randomly (ablation) | API-Net · CUB-200-2011 | accuracy: 86.4 -> 88.6 | Helped |
| use similar rather than dissimilar inter-class pairs (ablation) | API-Net pair construction · CUB-200-2011 | accuracy: 85.4 -> 87.2 | Helped |
| include both intra- and inter-class pairs rather than only one pair type (ablation) | API-Net pair construction · CUB-200-2011 | accuracy: 87.6 -> 88.6 | Helped |
| increase the class batch size from 10 to 30 (ablation) | API-Net pair construction · CUB-200-2011 | accuracy: 83.5 -> 88.6 | Helped |
| increase the number of sampled images per class from 2 to 4 (ablation) | API-Net pair construction · CUB-200-2011 | accuracy: 88.1 -> 88.6 | Helped |
| use the best API-Net configuration rather than the strongest listed baseline | fine-grained classification benchmarks · Aircraft | accuracy: 92.0 -> 93.9 | Helped |
| use the best API-Net configuration rather than the strongest listed baseline | fine-grained classification benchmarks · Stanford Cars | accuracy: 93.9 -> 95.3 | Helped |
Closest papers by meaning
- Channel Interaction Networks for Fine-Grained Image Categorization (AAAI 2020 · similarity 0.88)
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.85)
- Generalizable Whole Slide Image Classification with Fine-Grained Visual-Semantic Interaction (CVPR 2024 · similarity 0.84)
- From Coarse to Fine-Grained Open-Set Recognition (CVPR 2024 · similarity 0.84)
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 2024 · similarity 0.84)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.84)
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.