Research map / Label-efficient and robust learning / Recognition and long-tail
Learning a Mixture of Granularity-Specific Experts for Fine-Grained Categorization
Lianbo Zhang, Shaoli Huang, Wei Liu, Dacheng Tao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 KL-divergence constraint between experts (ablation) | two-expert MGE-CNN (ResNet-50) · CUB-200-2011 | top-1 accuracy: 87.9 -> 88.2 | Improved |
| add a second expert (ablation) | MGE-CNN (ResNet-50) · CUB-200-2011 | top-1 accuracy: 86.8 -> 88.2 | Improved |
| add a third expert (ablation) | MGE-CNN (ResNet-50) · CUB-200-2011 | top-1 accuracy: 88.2 -> 88.5 | Improved |
| combine second expert with first (ensemble) (ablation) | MGE-CNN (ResNet-50) · CUB-200-2011 | top-1 accuracy: 87.3 -> 87.9 | Improved |
| add GMP small-part branch to GAP branch (concat) (ablation) | first expert (ResNet-50) · CUB-200-2011 | top-1 accuracy: 85.4 -> 86.8 | Improved |
| add GMP small-part branch to GAP branch (concat) (ablation) | third expert (ResNet-50) · CUB-200-2011 | top-1 accuracy: 85.2 -> 86.1 | Improved |
| use GMP branch alone instead of GAP (ablation) | first expert (ResNet-50) · CUB-200-2011 | top-1 accuracy: 85.4 -> 83.8 | Hurt |
| raise attention threshold from 0.2 to 0.3 (ablation) | two-expert MGE-CNN · CUB-200-2011 | top-1 accuracy: 88.19 -> 88.44 | No clear effect |
| use ResNet-101 backbone instead of ResNet-50 (ablation) | MGE-CNN (3 experts) · CUB-200-2011 | top-1 accuracy: 88.5 -> 89.4 | Improved |
| use ResNet-101 backbone instead of ResNet-50 (ablation) | MGE-CNN · Stanford Cars | top-1 accuracy: 93.9 -> 93.6 | No clear effect |
| add MGE-CNN to ResNet-50 | ResNet-50 · CUB-200-2011 | top-1 accuracy: 85.4 -> 88.5 | Improved |
| add MGE-CNN to ResNet-50 | ResNet-50 · Stanford Cars | top-1 accuracy: 91.7 -> 93.9 | Improved |
Closest papers by meaning
- Bidirectional Logits Tree: Pursuing Granularity Reconcilement in Fine-Grained Classification (AAAI 2025 · similarity 0.84)
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.84)
- Scaling Laws for Fine-Grained Mixture of Experts (ICML 2024 · similarity 0.84)
- Learning to Specialize: Joint Gating-Expert Training for Adaptive MoEs in Decentralized Settings (NeurIPS 2025 · similarity 0.84)
- Multilinear Mixture of Experts: Scalable Expert Specialization through Factorization (NeurIPS 2024 · similarity 0.84)
- Fine-grained Classes and How to Find Them (ICML 2024 · similarity 0.83)
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.
