Awesome
openset-DA
This is an unofficial pytorch implementation of Open Set Domain Adaptation by Backpropagation.
Requirements
- Python 3.5+
- PyTorch 0.4
- torchvision
- scikit-learn
Usage
Run SVHN -> MNIST
python train.py --task s2m --gpu <gpu_id>
Run USPS -> MNIST
python train.py --task u2m --gpu <gpu_id> --grl-rampup-epochs 50
Run MNIST -> USPS
python train.py --task m2u --gpu <gpu_id> --grl-rampup-epochs 30
For more details and parameters, please refer to --help option.
Result
Please note that the parameters are not optimized.
SVHN -> MNIST
| OS | OS* | ALL | UNK | |
|---|---|---|---|---|
| This Code | 63.8 | 60.6 | 70.7 | 80.1 |
| Paper | 63.0 | 59.1 | 71.0 | 82.3 |
USPS -> MNIST
| OS | OS* | ALL | UNK | |
|---|---|---|---|---|
| This Code | 84.8 | 85.8 | 82.8 | 79.5 |
| Paper | 92.3 | 91.2 | 94.4 | 97.6 |
MNIST -> USPS
| OS | OS* | ALL | UNK | |
|---|---|---|---|---|
| This Code | 88.3 | 88.4 | 89.0 | 87.9 |
| Paper | 92.1 | 94.9 | 88.1 | 78.0 |