Revert "change conv model to use activation function and rerun a training run" as it degrades performance

change conv model to use activation function and rerun a training run

fix mdct (add square)

add non-pretrained-focus-liteNN model

fix vol4 function

add mean laplacian and mdct

add traditional method for comparison

remove error causeing (?) ResNET parameter and reduce range of lr

fix error in resnet code, don*t run constructor in place

readd cod modifications

Add simple fully convolutional network

added in_memory loading to reduce disk reads and increase speed

Add code so loss function uses torch.Size([x,1]) instead of torch.Size([x])

add testing notebook for convolutional neural network

add FNN with MSE loss

add missing changes from original repository

cleanup to make ready as submodule