Measuring the Interpretability of Unsupervised Representations via Quantized Reversed Probing

Open paper
Benchmark Model Rank Results
image-classification-on-objectnetDeepCluster(v2) (reverse linear probing)–Top-1 Accuracy: 19.73Top-5 Accuracy: 46.81
image-classification-on-objectnetMoCHi (reverse linear probing)–Top-1 Accuracy: 12.64Top-5 Accuracy: 31.71
image-classification-on-objectnetMoCo(v2) (reverse linear probing)–Top-1 Accuracy: 12.67Top-5 Accuracy: 31.45
image-classification-on-objectnetOBoW (reverse linear probing)–Top-1 Accuracy: 12.23Top-5 Accuracy: 31.72
image-classification-on-objectnetSeLa(v2) (reverse linear probing)–Top-1 Accuracy: 20.61Top-5 Accuracy: 48.83
image-classification-on-objectnetSwAV (reverse linear probing)–Top-1 Accuracy: 17.71Top-5 Accuracy: 43.64

Rank counts only results with a code link.