|
|
@ -22,9 +22,8 @@ A PyTorch implementation for paper *[Unsupervised Domain Adaptation by Backpropa |
|
|
|
- network structure proposed in the paper may be wrong for both 32x32 and 28x28 inputs |
|
|
|
- change last conv layer's filter to 4x4, get similar(actually higher) result |
|
|
|
- `GTSRBmodel()` |
|
|
|
- not successful |
|
|
|
- `AlexModel` |
|
|
|
- not successful, mainly due to the preprain model difference |
|
|
|
- not successful, mainly due to the pretrained model difference |
|
|
|
|
|
|
|
## Result |
|
|
|
|
|
|
|