Possible error with model implementation

[maxzuo]

Model taking too long to run on iPhone. Testing other models out there of larger sizes, the model should not take as long as it does to make a prediction.

Possible sources of problem:

• Model created in Keras not done properly (issue with architecture).
• Model implementation within Swift not optimized
• Model size too large.

[maxzuo]

Issue with architecture resolved by switching to almost an entirely new architecture.

• SRCNN went too deep too quickly. By adding more convolutional layers but gradually reaching 64 filters, the new model can help decrease the amount of time spent.
• New model's PSNR accuracy should also be improved.
• Further experimentation with PyTorch required. Model speed still not optimal.