Unsupervised Classification on ImageNet (val)
71.3AccuracyTrivialAugment
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TrivialAugmentnumber of epochs=400, batch size=2562022.02 | 71.3 | — | — | — | |
| TeachAugmentnumber of epochs=400, batch size=2562022.02 | 71 | — | — | — | |
| Baselinenumber of epochs=400, batch size=2562022.02 | 70.8 | — | — | — | |
| RandAugmentnumber of epochs=400, batch size=2562022.02 | 70.7 | — | — | — | |
| TeachAugmentnumber of epochs=200, batch size=2562022.02 | 70.2 | — | — | — | |
| Baselinenumber of epochs=200, batch size=2562022.02 | 70 | — | — | — | |
| RandAugmentnumber of epochs=200, batch size=2562022.02 | 70 | — | — | — | |
| TrivialAugmentnumber of epochs=200, batch size=2562022.02 | 68.7 | — | — | — | |
| TeachAugmentnumber of epochs=100, batch size=2562022.02 | 68.2 | — | — | — | |
| Baselinenumber of epochs=100, batch size=2562022.02 | 68.1 | — | — | — | |
| RandAugmentnumber of epochs=100, batch size=2562022.02 | 68 | — | — | — | |
| TrivialAugmentnumber of epochs=100, batch size=2562022.02 | 62.7 | — | — | — | |
| Self-ClassifierBackbone=ResNet-502021.03 | 41.1 | — | — | — | |
| Self-ClassifierBackbone=ResNet-50, Epochs=8002021.03 | 41.1 | 73.3 | 29.5 | 53.1 | |
| TWIST2021.10 | 40.6 | 74.3 | 30 | 57.7 | |
| Self-ClassifierBackbone=ResNet-50, Epochs=4002021.03 | 40.2 | 72.9 | 28.8 | 52.3 | |
| SCAN2021.10 | 39.9 | 72 | 27.5 | 51.2 | |
| SCANBackbone=ResNet-50, Pre-training=ImageNet-pretrained2021.03 | 39.9 | — | — | — | |
| SCANBackbone=ResNet-50, Epochs=800+1252021.03 | 39.9 | 72 | 27.5 | 51.2 | |
| Self-ClassifierBackbone=ResNet-50, Epochs=2002021.03 | 39.4 | 72.5 | 28.1 | 51.6 | |
| Self-ClassifierBackbone=ResNet-50, Epochs=1002021.03 | 37.3 | 71.2 | 26.1 | 49.2 | |
| BarlowTBackbone=ResNet-50, Evaluation Protocol=k-means classifier2021.03 | 34.2 | — | — | — | |
| BarlowTBackbone=ResNet-50, Epochs=1000, Evaluation protocol=fitting a k-means classifier2021.03 | 34.2 | 67.1 | 17.6 | 43.6 | |
| InfoMinBackbone=ResNet-50, Epochs=800, Evaluation protocol=fitting a k-means classifier2021.03 | 33.2 | 68.8 | 14.7 | 48.3 | |
| OBoWBackbone=ResNet-50, Evaluation Protocol=k-means classifier2021.03 | 31.1 | — | — | — | |
| OBOWBackbone=ResNet-50, Epochs=200, Evaluation protocol=fitting a k-means classifier2021.03 | 31.1 | 66.5 | 16.9 | 42 | |
| DINOBackbone=ResNet-50, Evaluation Protocol=k-means classifier2021.03 | 30.7 | — | — | — | |
| DINOBackbone=ResNet-50, Epochs=800, Evaluation protocol=fitting a k-means classifier2021.03 | 30.7 | 66.2 | 15.6 | 42.3 | |
| MoCoV2Backbone=ResNet-50, Evaluation Protocol=k-means classifier2021.03 | 30.6 | — | — | — | |
| MoCoV2Backbone=ResNet-50, Epochs=800, Evaluation protocol=fitting a k-means classifier2021.03 | 30.6 | 66.6 | 12 | 45.3 | |
| SeLaBackbone=ResNet-50, Pre-training=ImageNet-pretrained2021.03 | 30.5 | — | — | — | |
| SeLaBackbone=ResNet-50, Epochs=2802021.03 | 30.5 | 65.7 | 16.2 | 42 | |
| SwAVBackbone=ResNet-50, Evaluation Protocol=k-means classifier2021.03 | 28.1 | — | — | — | |
| SwAVBackbone=ResNet-50, Epochs=800, Evaluation protocol=fitting a k-means classifier2021.03 | 28.1 | 64.1 | 13.4 | 38.8 | |
| SimSiamBackbone=ResNet-50, Epochs=1002021.03 | 24.9 | 62.2 | 11.6 | 34.9 | |
| SimCLRv2Backbone=ResNet-50, Epochs=1000, Evaluation protocol=fitting a k-means classifier2021.03 | 22.4 | 61.5 | 11 | 34.9 | |
| SeLa2021.10 | — | 65.7 | 16.2 | 42 | |
| SelfClassifier2021.10 | — | 64.7 | 13.2 | 46.2 |