Making Art with GANs : Week 3 Assignment

-

I tried the Style transfer notebook with different images but I didn't like the results. I updated the iterations but I don't find the end product as interesting. Also, most the images are affected by style (colors) which I find a little limiting. I was able to troubleshoot every issue I had and took notes on this blog that I can reuse in case I were to need to run this notebook again. My goal is to work with video.

Style Transfer Notebook: https://colab.research.google.com/github/dvschultz/artificial-images/blob/master/neural_style_tf.ipynb


content image 1

Photo by me
Style 1
                                                    Grainstacks at the End of Summer, Morning Effect Painting by Claude Monet
!python neural_style.py \
--content_img /content/test1.jpeg\
--style_imgs /content/mon.jpg \
--max_iterations 300 \
--style_scale 0.8

Result 1



--------------------------------------------------------
Content image 2 

photo: me
Style image 2

Credit: kind-and-curious-ZDUXvlyU_iI-unsplash

RESULT 3

!python neural_style.py \
--content_img /content/sunset.png\
--style_imgs /content/kind.jpg \
--max_iterations 200 \
--style_scale 0.145



RESULT 3B
!python neural_style.py \
--content_img /content/sunset.png\
--style_imgs /content/kind.jpg \
--max_iterations 350 \
--style_scale 0.8



Content Image. 4 Style 4





-------------------------------

Content Image + Style image

Sol Lewitt would cry.


Result





--------------------------------------------
  • First attempt: Running the notebook without changes. Both images are the same size.

Content Image 

Credit: ganapathy-kumar-unsplash

Style image


Output


I'm not pleased with the image, but I am happy the code run.

  • Second attempt : Derrick suggested to make changes on the iterations (between 350-500)
 and to adjust the scale of the transfer image. Initially I got an error because of a typo
!python neural_style.py \
 --content_img /content/cont.jpg \
 --style_imgs /content/style.png \
 --max_iterations 500 \
 --style_scale 0.5



The image below I updated it with 300 itirerations scale 0.125


 Comparison video between Neural-Style-TF and Neural-Style-PT









    STEPS

First I picked two images one for content and one for the style . Then resizing the images to make sure the notebook will run smoothly. Currently I'm using the free Google Collab version, so using an ultra high res image  does not make sense. Like Derrick did in class I resize mine to make them 800 px on the largest side.

This notebook uses tensorflow 1 so the output should be 1.15.12

Adding a new line of code to check the GPU that will run the notebook

CODE :
!nvidia-smi -L

RESULT : GPU 0: Tesla K80 (UUID: GPU-ad28731c-79cf-6a4c-20e7-8a37c80a5d10
Location of folders




Popular posts from this blog

Socially Engaged Art and Digital Practice week 9 - Project 2

-
How do physical presence function in a networked world?  The physical presence is often overseen when thinking of the networked world we live in. Flavors, smells, physical touch are all part of our daily lives in a metropolitan city. Community engagement is what helps social networks thrive. During Covid some online communities flourished . Could this possibly be  a result of  a decade and a half of using social media platforms and smartphones? To me the networked world will relay on the human social experience of proximity. For example influencers rely on their following to consume things outside of the virtual interaction but in the networked world the followers get in close proximity through consumer products, or even places that they were referred to by influencers.  The symbiotic relationship between physical and online experience will continue to feed off each other in the near future.  Create something that addresses the materiality and/or Web3 reading (...
Read more

Socially Engaged Art and Digital Practice : week 5 Materials Presentation

-
Image
Materials group : Presentation    Angie, Cheer, Divya, Josephine, Lorena, Oscar, Spencer, Yvonne  Notes from  Digital Lives : Chapter 2 Production needed for digital lives Thinking how digital cultures are possible, ongoing human adaptations Evolution of internet, stages overlap and partially incorporate earlier stages 1962 J.C.R Licker “galactic networks” 1989 world wide web 70’ & 80’s military academic internet DARPA (Defense advanced Research Projects Agency) created DARPANET to facilitate communications. The internet began as a top-secret secured network 80’s scientific internet 90’s early National Science Foundation received permission to use ARPANET, a connection between scientists and the military. Scientists ->Universities -> social science, Art,  90’s counterculture avant-garde California based, social critique “geeks”, anti establishment, open-source Easy to use browsers 90’s mid emergent public internet -Commercial internet -Domesticated int...
Read more

Socially Engaged Art and Digital Practice : week 3

-
What is the role of bias in creating technological tools? I think using technological tools is one of the easiest ways to access a larger demographic therefore it should be created by All TYPES of people. For the past 20 years we have become more and more dependent on the information we consumed and exchange in a tech environment. Bias is real and it wasn't until less than a decade ago that the conversation about bias and the impact on people life's started to become more of a central topic. Now more than ever we need to give tools to people that otherwise don't have a voice in development of said technologies. Also understanding that Bias cannot disappear from one day to another, it a long process and we need to keep the conversation and education going.
Read more