Research map / Label-efficient and robust learning / Test-time adaptation
Test-Time Amendment with a Coarse Classifier 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 |
|---|---|---|---|
| Combine independently trained coarse and fine predictions by multiplying their probabilities | ResNet-18 classifiers · iNaturalist-19 | Top-1 error (%): 36.44 ± 0.061 -> 35.33 ± 0.037 | Helped |
| Combine independently trained coarse and fine predictions by multiplying their probabilities | ResNet-18 classifiers · iNaturalist-19 | Mistake severity: 2.39 ± 0.007 -> 2.15 ± 0.003 | Helped |
| Combine independently trained coarse and fine predictions by multiplying their probabilities | ResNet-18 classifiers · tieredImageNet-H | Top-1 error (%): 30.64 ± 0.030 -> 29.81 ± 0.081 | Helped |
| Combine independently trained coarse and fine predictions by multiplying their probabilities | ResNet-18 classifiers · tieredImageNet-H | Mistake severity: 7.07 ± 0.010 -> 6.95 ± 0.013 | Helped |
| Use HiE with MoCo-H | MoCo-H · iNaturalist-19, 10 images/label | Top-1 accuracy: 51.23 -> 55.13 | Helped |
| Use HiE with CrossEntropy | CrossEntropy · iNaturalist-19, 100 images/label | Hierarchical Distance@1: 1.17 -> 0.89 | Helped |
| Replace the independently trained coarse classifier with coarse predictions marginalized from the fine model (ablation) | HiE-Self · iNaturalist-19 | Top-1 error (%): 36.44 ± 0.061 -> 36.36 ± 0.047 | No clear effect |
| Add a hierarchical loss to marginalized coarse predictions (ablation) | Cross-Entropy-H · iNaturalist-19 | Top-1 error (%): 36.44 ± 0.061 -> 37.63 ± 0.054 | Hurt |
| Use hierarchy levels {6, 7} rather than leaf level {7} alone (ablation) | HiE · iNaturalist-19 species | Top-1 accuracy: 63.63 -> 64.65 | Helped |
| Add level 5 to levels {6, 7} (ablation) | HiE · iNaturalist-19 species | Top-1 accuracy: 64.65 -> 64.72 | Helped |
| Add level 4 to levels {5 - 7} (ablation) | HiE · iNaturalist-19 species | Top-1 accuracy: 64.72 -> 64.48 | Hurt |
| Add level 3 to levels {4 - 7} (ablation) | HiE · iNaturalist-19 species | Top-1 accuracy: 64.48 -> 64.47 | No clear effect |
Closest papers by meaning
- Fine-grained Classes and How to Find Them (ICML 2024 · similarity 0.85)
- Understanding Overadaptation in Supervised Fine-Tuning: The Role of Ensemble Methods (ICML 2025 · similarity 0.83)
- Bidirectional Logits Tree: Pursuing Granularity Reconcilement in Fine-Grained Classification (AAAI 2025 · similarity 0.83)
- Twofold Debiasing Enhances Fine-Grained Learning with Coarse Labels (AAAI 2025 · similarity 0.83)
- From Coarse to Fine-Grained Open-Set Recognition (CVPR 2024 · similarity 0.83)
- Fine-Tuning is Fine, if Calibrated (NeurIPS 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.