RameenAbdal / StyleFlow

StyleFlow: Attribute-conditioned Exploration of StyleGAN-generated Images using Conditional Continuous Normalizing Flows (ACM TOG 2021)
http://rameenabdal.github.io/StyleFlow
2.42k stars 342 forks source link

A Lot of General UI Questions #109

Open cyyeung1234 opened 3 years ago

cyyeung1234 commented 3 years ago

First off, I want to share my appreciation for the author's work. I think something truly groundbreaking is done here in this work.

However, I think overall the github and readme could use significant documentation improvement. Having successfully followed the setup instructions and played with the UI. I have several questions and comments.

1) What is the difference between main.py and main_attribute.py? This is not explained. 2) What is the purpose of the right image panel in main_attribute.py? 3) It looks like 4 buttons and icons do not do anything (add_new_document.png, open.png, paint_can.png, undo.png). It looks like only foot2.png and reset1.png are configured. Noting this in the readme would be helpful to others. 4) Are the selectable distribution of images (on UI start) real or generated? Do the location of the dots have any significance? 5) The paper mentions being able to do attribute-conditioned sampling. Any recommendations on how to perform this, even without the UI? 6) As others have noted, it looks like a tutorial needs to be added for photo opening. I've seen several YouTube videos with the Elon Musk example: https://www.youtube.com/embed/Lt4Z5oOAeEY. None of the default "dots" include the Elon Musk photo, so this must be an example of photo opening and real-photo editing. It would be great if a link to this example was provided, or if the updated tutorial included this demonstration!

marcelpadilla commented 2 years ago

I really want to know how to add an image myself! I got it running on windows 10 but the buttons on the left don't help me like you mentioned.