Research map / Label-efficient and robust learning / Recognition and long-tail
Grafit: Learning Fine-Grained Image Representations With Coarse Labels
Hugo Touvron, Alexandre Sablayrolles, Matthijs Douze, Matthieu Cord, Hervé Jégou
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 instance loss to kNN loss (FC head) (ablation) | Grafit FC · ImageNet-1k | mAP (%): 30.7 -> 41.7 | Improved |
| add instance loss to kNN loss (FC head) (ablation) | Grafit FC · CIFAR-100 | kNN top-1 accuracy (%): 70.0 -> 75.6 | Improved |
| add instance loss to cross-entropy (ablation) | Baseline · ImageNet-1k | kNN top-1 accuracy (%): 54.7 -> 65.0 | Improved |
| add coarse conditioning (ablation) | Grafit FC · ImageNet-1k | mAP (%): 41.7 -> 44.4 | Improved |
| add coarse conditioning (ablation) | Grafit FC · CIFAR-100 | mAP (%): 53.6 -> 55.0 | Improved |
| add coarse conditioning (ablation) | Baseline (cross-entropy) · ImageNet-1k | mAP (%): 22.7 -> 24.4 | Improved |
| use MLP projector instead of FC (ablation) | Grafit · CIFAR-100 | kNN top-1 accuracy (%): 75.6 -> 77.7 | Improved |
| use MLP projector instead of FC (ablation) | Grafit · ImageNet-1k | mAP (%): 44.4 -> 42.9 | Hurt |
| use oracle coarse labels at inference (ablation) | Grafit · ImageNet-1k | mAP (%): 42.9 -> 47.2 | Improved |
| add ClusterFit+ on top of baseline (ablation) | ClusterFit+ · ImageNet-1k | mAP (%): 22.7 -> 12.7 | Hurt |
| train Grafit with fine labels instead of coarse labels | Grafit · ImageNet-1k | kNN top-1 accuracy (%): 69.1 -> 79.6 | Improved |
| replace baseline with Grafit for transfer | Grafit · iNaturalist 2019 | top-1 accuracy (%): 73.7 -> 75.9 | Improved |
Later papers that compared with this method
2 later accepted papers report Grafit as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Learning Separable Fine-Grained Representation via Dendrogram Construction from Coarse Labels for Fine-grained Visual Recognition (ICCV 2025 · better on 2 of 2 shared results)
- MaskCon: Masked Contrastive Learning for Coarse-Labelled Dataset (CVPR 2023 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Coarse-to-Fine: Learning Compact Discriminative Representation for Single-Stage Image Retrieval (ICCV 2023 · similarity 0.89)
- Language-driven Fine-grained Retrieval (CVPR 2026 · similarity 0.88)
- A Weakly Supervised Fine Label Classifier Enhanced by Coarse Supervision (ICCV 2019 · similarity 0.88)
- Fine-grained Classes and How to Find Them (ICML 2024 · similarity 0.87)
- Embedding Label Structures for Fine-Grained Feature Representation (CVPR 2016 · similarity 0.87)
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.85)
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.
