Image Classification on ImageNet 10% label fraction 2012 (val)
76.2Top-1 AccSimMatchV2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SimMatchV2Pre-training Algorithm=None, Semi-supervised Algorithm=SimMatchV2 (Ours), Semi-supervised Epochs=~300, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 76.2 | 92.2 | |
| PAWSPre-training Algorithm=PAWS, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~300, Semi-supervised Epochs=~50, Parameters train=36.1M, Parameters inference=29.8M, Backbone=ResNet-502023.08 | 75.5 | — | |
| SimMatchV2Pre-training Algorithm=None, Semi-supervised Algorithm=SimMatchV2 (Ours), Semi-supervised Epochs=~100, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 74.8 | 91.7 | |
| SimMatchPre-training Algorithm=None, Semi-supervised Algorithm=SimMatch, Semi-supervised Epochs=~400, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 74.4 | 91.7 | |
| MoCo V2 (FixMatch-EMAN)Pre-training Algorithm=MoCo V2, Semi-supervised Algorithm=FixMatch-EMAN, Pre-training Epochs=~800, Semi-supervised Epochs=~300, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 73.9 | 91 | |
| CoMatchPre-training Algorithm=None, Semi-supervised Algorithm=CoMatch, Semi-supervised Epochs=~400, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 73.6 | 91.6 | |
| FixMatch-EMANPre-training Algorithm=None, Semi-supervised Algorithm=FixMatch-EMAN, Semi-supervised Epochs=~300, Parameters train=25.6M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 72.8 | 90.3 | |
| SSCLPre-training Algorithm=None, Semi-supervised Algorithm=SSCL, Semi-supervised Epochs=~800, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 72.1 | 90.9 | |
| WCLPre-training Algorithm=WCL, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~800, Semi-supervised Epochs=~60, Parameters train=32.2M, Parameters inference=29.8M, Backbone=ResNet-502023.08 | 72 | 91.2 | |
| FixMatchPre-training Algorithm=None, Semi-supervised Algorithm=FixMatch, Semi-supervised Epochs=~300, Parameters train=25.6M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 71.5 | 89.1 | |
| SWAVPre-training Algorithm=SWAV, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~800, Semi-supervised Epochs=~20, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 70.2 | 89.9 | |
| UDAPre-training Algorithm=None, Semi-supervised Algorithm=UDA, Semi-supervised Epochs=~530, Parameters train=25.6M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 68.8 | 88.8 | |
| BYOLPre-training Algorithm=BYOL, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~1000, Semi-supervised Epochs=~50, Parameters train=35.1M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 68.8 | 89 | |
| SimCLR V2Pre-training Algorithm=SimCLR V2, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~800, Semi-supervised Epochs=~60, Parameters train=32.2M, Parameters inference=29.8M, Backbone=ResNet-502023.08 | 68.4 | 89.2 | |
| MoCo V2 (Fine-tune)Pre-training Algorithm=MoCo V2, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~800, Semi-supervised Epochs=~20, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 66.1 | 87.9 | |
| SimCLRPre-training Algorithm=SimCLR, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~1000, Semi-supervised Epochs=~60, Parameters train=28.0M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 65.6 | 87.8 | |
| S4L-RotationPre-training Algorithm=None, Semi-supervised Algorithm=S4L-Rotation, Semi-supervised Epochs=~200, Parameters train=25.6M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 53.4 | 83.8 | |
| VAT+EntMin.Pre-training Algorithm=None, Semi-supervised Algorithm=VAT+EntMin., Semi-supervised Epochs=~100, Parameters train=25.6M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | 47 | 83.4 | |
| PCLPre-training Algorithm=PCL, Semi-supervised Algorithm=Fine-tune, Pre-training Epochs=~200, Semi-supervised Epochs=~20, Parameters train=23.8M, Parameters inference=25.6M, Backbone=ResNet-502023.08 | — | 85.6 |