neural-style-transfer | Generate novel artistic images using neural style | Computer Vision library

My end-goal is to create a script for neural style transfer, however, during writing code for said task, I stumbled upon a certain problem: the texture synthesis part of the algorithm seemed to have some problems with reproducing the artistic style. In order to solve this, I decided to create another script where I'd try to solve the task of texture synthesis using a neural network on its own.

TL;DR ... even after tackling the problem on its own, my script still produced blocky / noisy, non-sensible output.

I've tried having a look at how other people have solved this task, but most of what I found were more sophisticated solutions ("fast neural-style-transfer", etc.). Also, I couldn't find too many PyTorch implementations.

Since I've already spent the past couple of days on trying to fix this issue and considering that I'm new to the PyTorch-Framework, I have decided to ask the StackOverflow community for help and advice.

I use the VGG16 network for my model ...

I've been trying to recreate the work done in this blog post. The writeup is very comprehensive and code is shared via a collab.

What I'm trying to do is extract layers from the pretrained VGG19 network and create a new network with these layers as the outputs. However, when I assemble the new network, it highly resembles the VGG19 network and seems to contain layers that I didn't extract. An example is below.

Following this tutorial: https://www.digitalocean.com/community/tutorials/how-to-perform-neural-style-transfer-with-python-3-and-pytorch#step-2-%E2%80%94-running-your-first-style-transfer-experiment

When I run the example in Jupyter notebook, I get the following:

So, I've tried troubleshooting, which eventually got me to running it as per the github example (https://github.com/zhanghang1989/PyTorch-Multi-Style-Transfer) says to via command line: