- With help of neural style transfer, content image and style image generate new artistic image.
- Hidden layer activations individually play important role in deep learning applications.
- Accurate value of hyper parameter depends on specific application. Sometimes just reducing cost by increasing number of iterations makes no sense. This is neural style transfer is perfect application of this as image at 1000 iterations is same as image at 2000 iterations.
- Using pre-trained networks with transfer learning can save lot of time as well as computation cost.
Check result section in Project_report