Research map / Label-efficient and robust learning / Domain adaptation and generalization
Large Scale Fine-Grained Categorization and Domain-Specific Transfer Learning
Yin Cui, Yang Song, Chen Sun, Andrew Howard, Serge Belongie
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 |
|---|---|---|---|
| Increase Inception-v3 input resolution from 299 to 560 (ablation) | Inception-v3 on iNat · iNat minival | top-1 error rate (%): 29.93 -> 25.37 | Improved |
| Add fine-tuning on a more balanced subset (ablation) | Inception-v4 560 on iNat · iNat minival, head categories | top-1 error rate (%): 19.28 -> 17.33 | Improved |
| Add fine-tuning on a more balanced subset (ablation) | Inception-v4 560 on iNat · iNat minival, tail categories | top-1 error rate (%): 29.89 -> 24.15 | Improved |
| Add fine-tuning on a more balanced subset (ablation) | Inception-v3 560 on iNat · iNat minival | top-1 error rate (%): 25.4 (+ 4.5) -> 22.7 (+ 2.7) | Improved |
| Replace Inception-v3 with Inception-v4 (ablation) | iNat model with balanced-subset fine-tuning · iNat minival | top-1 error rate (%): 22.7 (+ 2.7) -> 20.8 (+ 1.9) | Improved |
| Add 12-crop inference (ablation) | Inception-v4 560 FT · iNat minival | top-1 error rate (%): 20.8 (+ 1.9) -> 19.2 (+ 1.6) | Improved |
| Ensemble Inception-v4 and Inception-ResNet-v2 models (ablation) | iNat models with 12-crop inference · iNat minival | top-1 error rate (%): 19.2 (+ 1.6) -> 18.1 (+ 1.1) | Improved |
| Pre-train on the domain-similarity-selected Subset B instead of ImageNet | Inception-v3 transfer learning · CUB200 | top-1 accuracy: 82.84 -> 88.76 | Improved |
| Pre-train on the domain-similarity-selected Subset B instead of ImageNet | Inception-v3 transfer learning · Stanford Dogs | top-1 accuracy: 84.19 -> 85.23 | Improved |
| Pre-train on the domain-similarity-selected source subset | Inception-v3 transfer learning · Food101 | top-1 accuracy: 88.65 -> 88.37 | No clear effect |
| Train on combined ImageNet + iNat rather than only one source domain (ablation) | ResNet-50 pre-training · ImageNet val | top-1 error rate: 24.33 -> 25.23 | Hurt |
| Train on combined ImageNet + iNat rather than only one source domain (ablation) | Inception-v3 pre-training · iNaturalist minival | top-1 error rate: 31.18 -> 30.29 | Improved |
Closest papers by meaning
- Data-free Knowledge Distillation for Fine-grained Visual Categorization (ICCV 2023 · similarity 0.88)
- An Erudite Fine-Grained Visual Classification Model (CVPR 2023 · similarity 0.88)
- Webly-Supervised Fine-Grained Visual Categorization via Deep Domain Adaptation (TPAMI 2016 · similarity 0.87)
- Novel Class Discovery for Ultra-Fine-Grained Visual Categorization (CVPR 2024 · similarity 0.87)
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.87)
- Advancing Fine-Grained Classification by Structure and Subject Preserving Augmentation (NeurIPS 2024 · 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.
