Skip to content
Atharva Patil edited this page Oct 22, 2019 · 11 revisions

Homework:

  • Watch these Videos:

  • Coding:

    • Train your own fast style transfer model and run the model in the browser with ml5.js
    • Or use any of the pre-trained Style Transfer models to create a new sketch
    • Publish your project on GitHub or your own blog, or record a video and put it on your blog. Add your project link below.

Some issues you might have when training the model with spell

  • During step 1 preparing your environment, after you add your style image to the style folder, you need to commit the changes to git, so the remote spell machine can get access to your changes.

    $ git add images ckpt
    $ git commit -m "Added required folders and images"
    
  • When you are on step 3 Training with style.py, you need to choose a GPU machine type, CPU machine wouldn't work.

    --machine-type k80
    
  • On step 4 Converting the model to ml5js, you need to install TensorFlow on your local computer. See more instruction here. Or you can run these commands on Spell, so you don't need to install TensorFlow locally, after you are done, spell cp your model back to your local computer.

  • Remember to replace things like 'YOUR_RUN_NUMBER', 'YOUR_FOLDER_NAME', 'YOUR_STYLE_IMAGE_NAME' to your own number and names.

See demos live

Resource:

Questions?

  • add your question here

Add your name / link to your assignment below

Clone this wiki locally