MIM-Refiner: A Contrastive Learning Boost from Intermediate Pre-Trained Representations

Open paper
Benchmark Model Rank Results
image-clustering-on-imagenetMIM-Refiner (D2V2-ViT-H/14)#2Accuracy: 67.3NMI: 87.2ARI: 42.2
image-clustering-on-imagenetMIM-Refiner (MAE-ViT-H/14)#4Accuracy: 64.6NMI: 85.3ARI: 45.5
self-supervised-image-classification-on-imagenetMIM-Refiner (D2V2-ViT-H/14)#5Top 1 Accuracy: 84.7%Number of Params: 632M
self-supervised-image-classification-on-imagenetMIM-Refiner (MAE-ViT-2B/14)#7Top 1 Accuracy: 84.5%Number of Params: 1890M
self-supervised-image-classification-on-imagenetMIM-Refiner (MAE-ViT-H/14#8Top 1 Accuracy: 83.7%Number of Params: 632M
self-supervised-image-classification-on-imagenetMIM-Refiner (D2V2-ViT-L/16)#9Top 1 Accuracy: 83.5%Number of Params: 307M
self-supervised-image-classification-on-imagenetMIM-Refiner (MAE-ViT-L/16)#10Top 1 Accuracy: 82.8%Number of Params: 307M

Rank counts only results with a code link.