Color transfer

[MichielStock]

No description provided.

[MichielStock]

Starting to look good.

• add your name and a title to the notebook
• maybe there is a more elegant way of downloading the images? You might check some of the pluto docs. Would be nice if there are buttons I can select my own images.
• best to split your functions into different cells
• you can represent every pixel as the RGB + location and use k-means clustering to get prototype pixels. The obtained centres can be mapped and used for prediction again. This way you don't need downscaling.
• When making your own types (Cluster) adding type annotation greatly improves performance
• you might tidy up the organization of your functions.

Bottom line: make sure you don't have to do OT on all the pixels. Make this part of your preprocessing.

This was my attempt (a long time ago): https://github.com/MichielStock/Teaching/blob/master/Optimal_transport/color_transfer.py

Below is the advice I got from Mathieu Blondel

Good luck!

[juhlee]

Thank you for the detailed comment!

[MichielStock]

It is shaping up to be a very impressive project. When uploading your notebook, I still have some trouble doing this right, mainly because the links to images are relative (better use hyperlinks to github). Maybe have someone else testing it that works locally fine.

[juhlee]

Thank you for an important comment...! I did not expect "image_load()" to not work properly.

[juhlee]

When first opening the notebook, did the two default sample images (cityscape.jpg, sunset.jpg) not load properly?

[MichielStock]

no, check it by loading the hyperlink to the notebook in Pluto, this is how people will use it.

…

On 26 Jan 2022, at 18:45, Ju Hyung Lee ***@***.***> wrote: When first opening the notebook, did the two default sample images (cityscape.jpg, sunset.jpg) not load properly? — Reply to this email directly, view it on GitHub <#29 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABN5PI5WLUHBELN7EZAXZ7DUYAXL3ANCNFSM5LE577UA>. You are receiving this because you authored the thread.

[juhlee]

Thank you!

• I used "try/catch" statement to first try loading locally, then download the image if an local loading did not work.
• I then confirmed the notebook to run fine in another colleague's (outside of STMO course) computer.

[vandeveldet]

Great notebook and very clear walkthrough. Nice to have options available and see the effect on the clustering/color transfer. Initial issues with package loading were resolved by an update by @juhlee. Feel free to add a function when taking a repetitive action. Some small typo's: Eucl_i_dian distance and function reassi_gn_ing(). It worked well when adding new figures for Patrick and Spongebob ;).