Bug fix: now default to train on GPU, fall back to CPU if needed.
Fully working adaptation model based on DANN architecture
Source dataset: MNIST
Target dataset: MNISTM
MNIST ACC: 99%
MNISTM ACC: 93.5% | Without Adaptation: 51%
Train:
python3 train.py -c config.json
Test:
python3 test.py -c config.json -r [checkpoint]