Test different fast transfer optimizations

[tupini07]

Find out where it is better to do the optimiztion step and see if the feature loss is actually better than the content loss

• optimization step after processing each step of image
• aggregate loss for whole batch and then do optimization step
• test if feature loss is better than content loss, or if they can be used jointly

[tupini07]

The fast neural transfer paper uses Adam as the optimizer.