Benchmark samples with Inception score.

Directory structure should be
/results
/{model_type}
/{run_name}
/models
/images
/bench

There's also a /results/ref directory which will store the ground truth results.
{model_type} differentiates between major differences in model i.e. different loss function. {run_name} differentiates between hyperparameters.

As mentioned in title, need to get the FID score from https://github.com/mseitzer/pytorch-fid working with the output of our models.

Save sample images in the respective directory i.e. for iteration i, save the sampled images to
/results/{model_type}/{run_name}/images/{batch_num}/{i}.png

Benchmark samples from training with FID scores.

Get SAGAN loss into a train_D and train_G function.

Write results to tensorboardX server for viewing (can save everything first and then display later).

Trainer class hasn't been implemented

Save losses after every epoch i.e. sum losses for D and G for every iteration within an epoch and save after that epoch.

Implement reloading of weights to continue training from a certain epoch (also need to update losses file).

Write custom loss function for each type of GAN in their respective definition files.

As above. Perform hyperparameter search over the same number of epochs (mainly from scores I guess)

As mentioned in the title, need to get the code from https://github.com/sbarratt/inception-score-pytorch working with the output of our models.