Pushing the limits of self-supervised ResNets: Can we outperform supervised learning without labels on ImageNet?

Benchmark Model Rank Results
image-classification-on-objectnetRELICv2#69Top-1 Accuracy: 25.9
image-classification-on-objectnetRELIC#73Top-1 Accuracy: 23.8
image-classification-on-objectnetBYOL#74Top-1 Accuracy: 23
image-classification-on-objectnetSimCLR#81Top-1 Accuracy: 14.6
self-supervised-image-classification-on-imagenetReLICv2 (ResNet-200 x2)#21Top 1 Accuracy: 80.6%Number of Params: 250M
self-supervised-image-classification-on-imagenetReLICv2 (ResNet200)#26Top 1 Accuracy: 79.8%Number of Params: 63M
self-supervised-image-classification-on-imagenetReLICv2 (ResNet-50 4x)#31Top 1 Accuracy: 79.4%Number of Params: 375M
self-supervised-image-classification-on-imagenetReLICv2 (ResNet152)#32Top 1 Accuracy: 79.3%Number of Params: 58M
self-supervised-image-classification-on-imagenetReLICv2 (ResNet-50 x2)#36Top 1 Accuracy: 79%Number of Params: 94M
self-supervised-image-classification-on-imagenetReLICv2 (ResNet101)#39Top 1 Accuracy: 78.7%Number of Params: 44M
self-supervised-image-classification-on-imagenetReLICv2 (ResNet-50)#51Top 1 Accuracy: 77.1%Number of Params: 25M
semantic-segmentation-on-cityscapes-valReLICv2#70mIoU: 75.2
semantic-segmentation-on-cityscapes-valBYOL#73mIoU: 74.6
semantic-segmentation-on-pascal-voc-2012-valReLICv2#12mIoU: 77.9%
semantic-segmentation-on-pascal-voc-2012-valDetCon#15mIoU: 77.3%
semantic-segmentation-on-pascal-voc-2012-valBYOL#18mIoU: 75.7%
semi-supervised-image-classification-on-1RELICv2#36Top 1 Accuracy: 58.1%Top 5 Accuracy: 81.3
semi-supervised-image-classification-on-2RELICv2 (ResNet-50)#33Top 1 Accuracy: 72.4%Top 5 Accuracy: 91.2%