models/research/domain_adaptation/pixel_domain_adaptation/baselines/README.md

The best baselines are obtainable via the following configuration:

MNIST => MNIST_M

Accuracy: MNIST-Train: 99.9 MNIST_M-Train: 63.9 MNIST_M-Valid: 63.9 MNIST_M-Test: 63.6

Learning Rate = 0.0001 Weight Decay = 0.0 Number of Steps: 105,000

MNIST => USPS

Accuracy: MNIST-Train: 100.0 USPS-Train: 82.8 USPS-Valid: 82.8 USPS-Test: 78.9

Learning Rate = 0.0001 Weight Decay = 0.0 Number of Steps: 22,000

MNIST_M => MNIST

Accuracy: MNIST_M-Train: 100 MNIST-Train: 98.5 MNIST-Valid: 98.5 MNIST-Test: 98.1

Learning Rate = 0.001 Weight Decay = 0.0 Number of Steps: 604,400

MNIST_M => MNIST_M

Accuracy: MNIST_M-Train: 100.0 MNIST_M-Valid: 96.6 MNIST_M-Test: 96.4

Learning Rate = 0.001 Weight Decay = 0.0 Number of Steps: 139,400

USPS => USPS

Accuracy: USPS-Train: 100.0 USPS-Valid: 100.0 USPS-Test: 96.5

Learning Rate = 0.001 Weight Decay = 0.0 Number of Steps: 67,000