Methodology

Self-Supervised Learning

Learning representations without labeled data.

1
Datasets
8
Results
top-1-accuracy
Canonical metric
Canonical Benchmark

ImageNet Linear Probe

Measures representation quality via linear classification on frozen features

Primary metric: top-1-accuracy
View full leaderboard

Top 10

Leading models on ImageNet Linear Probe.

RankModeltop1_accuracyYearSource
1
DINOv2 ViT-g/14
86.52026paper
2
DINOv2 ViT-g/14
86.52026paper
3
DINOv2 ViT-L/14
86.32026paper
4
CLIP ViT-L/14
85.32026paper
5
SimCLRv2 (ResNet-152 3x)
79.82026paper
6
MAE ViT-H/14
77.22026paper
7
MAE ViT-H/14
76.62026paper
8
MAE ViT-L/16
76.02026paper

All datasets

1 dataset tracked for this task.

ImageNet Linear Probe

CANONICAL
8results·top-1-accuracy
Top: DINOv2 ViT-g/14 — 86.5

Related tasks

Other tasks in Methodology.

Continual LearningFew-Shot LearningTransfer Learning