Many people may not find out about this parameter. In fact, for some people is relatively low memory, or who want quick results are very useful.
What is lightweight encoder?
The lightweight nature is to reduce the complexity of the neural network (What is a neural network? This ....) Enabling this option after the final results will decline.
Lightweight encoder what's the use?
- Acceleration time is reduced each iteration
- Low with low memory also run up.
How Lightweight encoder open?
Change the configuration only when the model first start can be set. Click after the jump trainxx.bat there is a black interface is called: Use lightweight autoenconder?
This option defaults to n, do not start.
If you need to, enter y to start.
Lightweight model enabled encoder running up and obviously a lot faster. This parameter H128 and H64 have, SAE has been abandoned.
SEA may be achieved by reducing the value of AutoEncoder dims and the Encoder dims per channel.