Janelle Shane @JanelleCShane Research Scientist in optics. Plays with neural networks. Avid reader, writer, and player of Irish flute. she/her. wandering.shop/@janellecshane Jul. 25, 2019 1 min read

I used @runwayml to reconstruct this scene from The Last Jedi by having DeepLab recognize all the objects in each frame, and then having SPADE try to paint the original frame from the objects.
 https://youtu.be/sEd1EO8eIfw 

More on the reconstruction process here:  https://aiweirdness.com/post/186539822077/the-world-through-the-eyes-of-a-neural-net 
This is the first step, where DeepLab segments an image into objects. "Lightsaber" is not a category it has, so it improvises.

Next, SPADE has to paint a picture just from DeepLab's labeled image. It sees person with a baseball bat, so it looks a lot like it tried to paint Luke in a baseball jersey.

Here's another scene from TLJ as segmented by DeepLab (which sees people with teddy bears hiding behind the rock) and SPADE (which dutifully has to draw the teddy bears)

I like this sequence. DeepLab didn't detect the lightsabers & decided the ground was snow. (And also a surfboard). Trying to recover the original image, SPADE looked at all the snow & decided to put Luke and Kylo in snowsuits.

I'm really excited to see what these will look like as segmentation & reconstruction get better. Could one segment a scene from Pride and Prejudice and replace all the "human" labels with "bear"? Or "refrigerator"?

Here's a cool variation where @jonathanfly gives SPADE not a DeepLab segmentation but frames from cartoons or 8-bit video games.

flawless reconstruction by our robot overlords

i love this!


You can follow @JanelleCShane.



Bookmark

____
Tip: mention @threader_app on a Twitter thread with the keyword “compile” to get a link to it.

Enjoy Threader? Sign up.

Threader is an independent project created by only two developers. The site gets 500,000+ visits a month and our iOS Twitter client was featured as an App of the Day by Apple. Running this space is expensive and time consuming. If you find Threader useful, please consider supporting us to make it a sustainable project.