Clustering on CIFAR10 small-scale (val)
95.1NMIProbing CLIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Probing CLIPBackbone=ViT-L/14, Evaluation protocol=Supervised Probing2023.03 | 95.1 | 98.1 | 95.8 | |
| TEMIPre-training type=natural language supervised (CLIP), Backbone=ViT-L/14, Pre-trained on external data=true, Evaluation protocol=Clustering2023.03 | 92.6 | 96.9 | 93.2 | |
| Probing DINOBackbone=ViT-B/16, Evaluation protocol=Supervised Probing2023.03 | 92.5 | 96.8 | 93.1 | |
| TEMIPre-training type=supervised, Backbone=ViT-L/16, Evaluation protocol=Clustering2023.03 | 91.8 | 96 | 91.6 | |
| Probing MSNBackbone=ViT-L/16, Evaluation protocol=Supervised Probing2023.03 | 91.5 | 96.4 | 92.3 | |
| Probing Sup.Backbone=ViT-L/16, Evaluation protocol=Supervised Probing, Pre-trained on external data=true2023.03 | 91.5 | 96.5 | 92.4 | |
| ProPosEvaluation protocol=Clustering, Includes additional dataset splits during training=true2023.03 | 88.6 | 94.3 | 88.4 | |
| TEMIPre-training type=DINO, Backbone=ViT-B/16, Evaluation protocol=Clustering2023.03 | 88.6 | 94.5 | 88.5 | |
| TSPEvaluation protocol=Clustering2023.03 | 88 | 94 | 87.5 | |
| SPICEEvaluation protocol=Clustering2023.03 | 86.5 | 92.6 | 85.2 | |
| TEMIPre-training type=MSN, Backbone=ViT-L/16, Pre-trained on external data=true, Evaluation protocol=Clustering2023.03 | 82.9 | 90 | 80.7 | |
| PCLEvaluation protocol=Clustering2023.03 | 80.2 | 87.4 | 76.6 | |
| SCANEvaluation protocol=Clustering2023.03 | 79.7 | 88.3 | 77.2 | |
| NNMEvaluation protocol=Clustering2023.03 | 74.8 | 84.3 | 70.9 | |
| PICAEvaluation protocol=Clustering2023.03 | 59.1 | 69.6 | 51.2 | |
| DCCMEvaluation protocol=Clustering2023.03 | 49.6 | 62.3 | 40.8 | |
| DACEvaluation protocol=Clustering2023.03 | 39.6 | 52.2 | 30.6 |