AutoEncoder with MNIST dataset
Requirement
- python 3.6
- pytorch 0.4.1
- torchvision
- visdom
- numpy
- matplotlib
Model preview
MLP AutoEncoder result
simple model
dimension reduction to 3d
3D scatter using matplotlib
compare three axis (-0.5 ~ 0.5)
Convolutional AutoEncoder result
simple model
noise reduction
Filter comparison
| simple model | noise reduction |
|---|---|
 |
 |
I can not find a dramatic difference because each model learned until convergence.