Experiments (Ivan)

ResNet50 c доменным классификатором как у ванила данн

Resnet50 classificator vanila_dann 141 layer freeze 200 epoch 10 steps per epoch 64 bs

Model	A → W	D→ W	W→ D	A→ D	D → A	W→ A
DANN Resnet Обзор	0.826	0.978	1	0.833	0.668	0.661
ResNet with domain loss	0.8451	0.9349	0.9754	0.7969	0.6719	0.6893
ResNet without domain loss	0.7604	0.9674	0.9911	0.7924	0.6403	0.6317

Model	A → W	D→ W	W→ D	A→ D	D → A	W→ A
DANN Resnet Обзор	0.826	0.978	1	0.833	0.668	0.661
ResNet 141 freeze with domain loss	0.71484	-	-	0.62708	0.63246	0.67543
ResNet 141 freeze without domain loss	0.77995	0.96875	0.99583	0.79583	0.63565	0.63175
ResNet 129 freeze without domain loss	0.7849	-	-	-	-	-
ResNet 72 freeze without domain loss	0.7348	-	-	-	-	-
ResNet 0 freeze without domain loss	0.5443	-	-	-	-	-

Описание обучения и модели(Experement 4): Resnet50 classificator simple 141 layer freeze 200 epoch 20 steps per epoch 32 bs

Model	A → W	D→ W	W→ D	A→ D	D → A	W→ A
DANN Resnet Обзор	0.826	0.978	1	0.833	0.668	0.661
ResNet with domain loss	0.71484	-	-	0.62708	0.63246	0.67543
ResNet without domain loss	0.77995	0.96875	0.99583	0.79583	0.63565	0.63175

Описание обучения и модели(Experement 5): Resnet50 classificator test5 141 layer freeze 200 epoch 20 steps per epoch 32 bs

Model	A → W	D→ W	W→ D	A→ D	D → A	W→ A
DANN Resnet Обзор	0.826	0.978	1	0.833	0.668	0.661
ResNet with domain loss	0.8503	0.9154	0.9263	0.7902	0.6708	0.7109

ADDA

Model	MNIST->USPS	USPS->MNIST	MNIST->MNISTM
source	0.8457	0.5877	0.5535
ADDA	0.9413	0.8205	0.6396