Big Transfer (BiT): General Visual Representation Learning

Open paper
Benchmark Model Rank Results
fine-grained-image-classification-on-oxford-102-flowersBiT-L (ResNet)#2Accuracy: 99.63%Top-1 Error Rate: 0.37
fine-grained-image-classification-on-oxford-102-flowersBiT-M (ResNet)#4Accuracy: 99.30%Top-1 Error Rate: 0.70
fine-grained-image-classification-on-oxford-iiit-petsBiT-L (ResNet)#2Accuracy: 96.62Top-1 Error Rate: 3.38%
fine-grained-image-classification-on-oxford-iiit-petsBiT-M (ResNet)#5Accuracy: 94.47Top-1 Error Rate: 5.53%
image-classification-on-vtab-1kBiT-L (50 hypers/task)#2Top-1 Accuracy: 78.72
image-classification-on-vtab-1kBiT-L#5Top-1 Accuracy: 76.3
image-classification-on-vtab-1kBiT-M#10Top-1 Accuracy: 70.6
image-classification-on-vtab-1kBiT-S#13Top-1 Accuracy: 66.9
image-classification-on-omnibenchmarkBiT-M#7Average Top-1 Accuracy: 40.4
image-classification-on-cifar-10BiT-L (ResNet)#8Percentage correct: 99.37
image-classification-on-cifar-10BiT-M (ResNet)#33Percentage correct: 98.91
image-classification-on-flowers-102BiT-L (ResNet)#10Accuracy: 99.63
image-classification-on-flowers-102BiT-M (ResNet)#15Accuracy: 99.30
image-classification-on-cifar-100BiT-L (ResNet)#11Percentage correct: 93.51
image-classification-on-cifar-100BiT-M (ResNet)#20Percentage correct: 92.17
image-classification-on-imagenet-realBiT-L#20Accuracy: 90.54%Params: 928M
image-classification-on-imagenet-realBiT-M#31Accuracy: 89.02%
image-classification-on-objectnetBiT-L (ResNet-152x4)#28Top-1 Accuracy: 58.7Top-5 Accuracy: 80
image-classification-on-objectnetBiT-M (ResNet-152x4)#39Top-1 Accuracy: 47.0Top-5 Accuracy: 69
image-classification-on-objectnetBiT-S (ResNet-152x4)#55Top-1 Accuracy: 36.0Top-5 Accuracy: 57
image-classification-on-imagenetBiT-L (ResNet)#86Top 1 Accuracy: 87.54%Top 5 Accuracy: 98.46
image-classification-on-imagenetBiT-M (ResNet)#252Top 1 Accuracy: 85.39%Number of params: 928M
object-detection-on-coco-val2017RetinaNet (BiT-L)#337box AP: 43.8

Rank counts only results with a code link.