Can We Evaluate Domain Adaptation Models Without Target-Domain Labels?

Requirements

pytorch>=1.8.0
torchvision>=0.8.2
Python 3.8+
timm

Dataset

Following datasets can be downloaded automatically into data folder:

• Office31
• OfficeHome
• VisDA2017
• DomainNet

Demo

To train a DAN method on Office31 D2A and save the model:

CUDA_VISIBLE_DEVICES=0 python dan.py data/office31 -d Office31 -s D -t A -a resnet50 --epochs 20 --seed 0 --log logs/dan/Office31_D2A

To calculate a DAN method's transfer score on Office31 D2A:

CUDA_VISIBLE_DEVICES=0 python dan.py data/office31 -d Office31 -s D -t A -a resnet50 --epochs 20 --seed 0 --log logs/dan/Office31_D2A --phase evaluation