Image Classification on Places 205-way (test)
62.7Top-1 AccuracySEER
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SEERArch.=RG256, Pre-training source=uncurated data, Evaluation Protocol=linear evaluation2021.03 | 62.7 | — | — | — | — | — | — | |
| SEERArch.=RG128, Pre-training source=uncurated data, Evaluation Protocol=linear evaluation2021.03 | 61.9 | — | — | — | — | — | — | |
| SwAVArch.=RG128, Pre-training source=ImageNet, Evaluation Protocol=linear evaluation2021.03 | 59.9 | — | — | — | — | — | — | |
| SwAVEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 56.7 | — | — | — | — | — | — | |
| SwAVArch.=RN50, Pre-training source=ImageNet, Evaluation Protocol=linear evaluation2021.03 | 56.7 | — | — | — | — | — | — | |
| SupervisedArch.=RG128, Pre-training source=ImageNet, Evaluation Protocol=linear evaluation2021.03 | 56 | — | — | — | — | — | — | |
| MOAM* full%-labels=10, backbone=ResNet50v2, network width=4x wider2019.05 | 54.2 | — | — | — | — | — | 83.3 | |
| MOAM* + pseudo label%-labels=10, backbone=ResNet50v2, network width=4x wider2019.05 | 54.2 | — | — | — | — | — | 83.3 | |
| Supervised%-labels=100, backbone=ResNet50v2, network width=4x wider2019.05 | 53.7 | — | — | — | — | — | 83.1 | |
| SimCLREvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 53.3 | — | — | — | — | — | — | |
| SupervisedEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 53.2 | — | — | — | — | — | — | |
| MoCov2Evaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 52.9 | — | — | — | — | — | — | |
| Supervised%-labels=100, backbone=ResNet50v22019.05 | 52.5 | — | — | — | — | — | 81.9 | |
| BoWNetEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 51.1 | — | — | — | — | — | — | |
| PIRLEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 49.8 | — | — | — | — | — | — | |
| PCLEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 49.8 | — | — | — | — | — | — | |
| MOAM*%-labels=10, backbone=ResNet50v2, network width=4x wider2019.05 | 49.5 | — | — | — | — | — | 79.2 | |
| Pseudolabels%-labels=10, backbone=ResNet50v22019.05 | 48.2 | — | — | — | — | — | 78.1 | |
| MoCoEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 46.9 | — | — | — | — | — | — | |
| S4L-Rotation%-labels=10, backbone=ResNet50v22019.05 | 46.6 | — | — | — | — | — | 76.4 | |
| NPID++Evaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 46.4 | — | — | — | — | — | — | |
| VAT + EntMin%-labels=10, backbone=ResNet50v22019.05 | 46.2 | — | — | — | — | — | 76.4 | |
| SS Rotation%-labels=10, fine-tuning=true, backbone=ResNet50v22019.05 | 45.9 | — | — | — | — | — | 75.4 | |
| SS Exemplar%-labels=10, fine-tuning=true, backbone=ResNet50v22019.05 | 45.9 | — | — | — | — | — | 75.6 | |
| S4L-Exemplar%-labels=10, backbone=ResNet50v22019.05 | 45.9 | — | — | — | — | — | 75.9 | |
| VAT%-labels=10, backbone=ResNet50v22019.05 | 45.8 | — | — | — | — | — | 76.4 | |
| RotNetEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 45 | — | — | — | — | — | — | |
| Supervised%-labels=10, backbone=ResNet50v22019.05 | 44.7 | — | — | — | — | — | 75 | |
| Pseudolabels%-labels=1, backbone=ResNet50v22019.05 | 41.8 | — | — | — | — | — | 71.6 | |
| SS Rotation+%-labels=0, transfer training epochs=520, backbone=ResNet50v22019.05 | 41.7 | — | — | — | — | — | 71.4 | |
| SS Exemplar+%-labels=0, transfer training epochs=520, backbone=ResNet50v22019.05 | 39.8 | — | — | — | — | — | 69 | |
| S4L-Rotation%-labels=1, backbone=ResNet50v22019.05 | 38 | — | — | — | — | — | 67.3 | |
| VAT + EntMin%-labels=1, backbone=ResNet50v22019.05 | 36.4 | — | — | — | — | — | 65.9 | |
| SS Rotation%-labels=1, fine-tuning=true, backbone=ResNet50v22019.05 | 36.3 | — | — | — | — | — | 66.1 | |
| Supervised%-labels=1, backbone=ResNet50v22019.05 | 36.2 | — | — | — | — | — | 65.4 | |
| VAT%-labels=1, backbone=ResNet50v22019.05 | 35.9 | — | — | — | — | — | 64.9 | |
| S4L-Exemplar%-labels=1, backbone=ResNet50v22019.05 | 32.2 | — | — | — | — | — | 61.2 | |
| SS Exemplar%-labels=1, fine-tuning=true, backbone=ResNet50v22019.05 | 31.1 | — | — | — | — | — | 60 | |
| ab -> L(cl)Classifier=Linear logistic regression, Feature state=Frozen, Loss=Classification loss, Task=Grayscale prediction2016.11 | — | 15.6 | 22.5 | 24.8 | 25.1 | 23 | — | |
| ab -> L(reg)Classifier=Linear logistic regression, Feature state=Frozen, Loss=L2 loss, Task=Grayscale prediction2016.11 | — | 15.9 | 22.8 | 25.6 | 26.2 | 24.9 | — | |
| Doersch et al.Classifier=Linear logistic regression, Feature state=Frozen, Pre-training dataset=ImageNet2016.11 | — | 19.7 | 26.7 | 31.9 | 32.7 | 30.9 | — | |
| Donahue et al.Classifier=Linear logistic regression, Feature state=Frozen, Pre-training dataset=ImageNet2016.11 | — | 22 | 28.7 | 31.8 | 31.3 | 29.7 | — | |
| GaussianClassifier=Linear logistic regression, Feature state=Frozen, Initialization=Random Gaussian2016.11 | — | 15.7 | 20.3 | 19.8 | 19.1 | 17.5 | — | |
| ImageNet-labelsClassifier=Linear logistic regression, Feature state=Frozen, Supervision=Supervised, Pre-training dataset=ImageNet2016.11 | — | 22.7 | 34.8 | 38.4 | 39.4 | 38.7 | — | |
| Krahenbuhl et al.Classifier=Linear logistic regression, Feature state=Frozen, Initialization=Stacked k-means2016.11 | — | 21.4 | 26.2 | 27.1 | 26.1 | 24 | — | |
| L -> ab(cl)Classifier=Linear logistic regression, Feature state=Frozen, Loss=Classification loss, Task=Colorization2016.11 | — | 16.4 | 27.5 | 31.4 | 32.1 | 30.2 | — | |
| L -> ab(reg)Classifier=Linear logistic regression, Feature state=Frozen, Loss=L2 loss, Task=Colorization2016.11 | — | 16.2 | 26.5 | 30 | 30.5 | 29.4 | — | |
| Noroozi & FavaroClassifier=Linear logistic regression, Feature state=Frozen, Backbone=AlexNet (stride 2 conv1)2016.11 | — | 23 | 32.1 | 35.5 | 34.8 | 31.3 | — | |
| Owens et al.Classifier=Linear logistic regression, Feature state=Frozen, Pre-training dataset=Other large-scale data2016.11 | — | 19.9 | 29.3 | 32.1 | 28.8 | 29.8 | — | |
| Pathak et al.Classifier=Linear logistic regression, Feature state=Frozen, Pre-training dataset=ImageNet2016.11 | — | 18.2 | 23.2 | 23.4 | 21.9 | 18.4 | — | |
| Places-labelsClassifier=Linear logistic regression, Feature state=Frozen, Supervision=Supervised, Pre-training dataset=Places2016.11 | — | 22.1 | 35.1 | 40.2 | 43.3 | 44.6 | — | |
| Split-Brain Auto (cl, cl)Classifier=Linear logistic regression, Feature state=Frozen, Backbone=AlexNet, Architecture=Split-Brain Autoencoder2016.11 | — | 21.3 | 30.7 | 34 | 34.1 | 32.5 | — | |
| Wang & GuptaClassifier=Linear logistic regression, Feature state=Frozen, Pre-training dataset=Other large-scale data2016.11 | — | 20.1 | 28.5 | 29.9 | 29.7 | 27.9 | — | |
| Zhang et al.Classifier=Linear logistic regression, Feature state=Frozen, Pre-training dataset=ImageNet2016.11 | — | 16 | 25.7 | 29.6 | 30.3 | 29.7 | — |