Just keeping this as a todo
There are additional ways to parametrize the model defintion for example define the number or encoder/decoder layers, channel dimension decrease rate (1 to keep the same n channel as of now, 2 to divide by a factor of 2 each time etc...).
ltetrel
commented
2 years ago
Just keeping this as a todo There are additional ways to parametrize the model defintion for example define the number or encoder/decoder layers, channel dimension decrease rate (1 to keep the same n channel as of now, 2 to divide by a factor of 2 each time etc...).
This allow for much better dyamic parametrization of the model, as an example: https://github.com/SIMEXP/DeepNeuroAN/blob/28750d81e046816dd1edbad1564344ed5ccd4dcd/deepneuroan/models.py#L547