Neural Networks 3D Simulation

A simple 3D simulation of a few models from a simple regression to a network with a convolutional layer.
I trained models with Keras and used MNIST handwritten digit dataset. Visualizations are made with Blender 2.8 (Python API).
Manim Python library is used for text animations.

[@nomadow2423]

Lovely animation. Thanks for sharing

[@AquarianTeacher]

Brilliant!🌟

[@MochiClips]

Loved it!

[@IvoVacca]

This is cool! I made the same using tensorflow and Blender.

[@danielkrei]

I checked it out :) very cool stuff!

[@fooboomoo]

Yea, this is pretty sweet :)

[@torque8047]

Really cool!

[@SalihSabir]

Wow dude. Bravo!

[@world_production]

Hello ! Can you make a tutorial on how to make this in blender ? Thanks 👍

[@danielkrei]

Hello and thank you for your interest! This is made with blender python API and Keras. I’ll clean the code and make a tutorial soon:)

[@world_production]

@@danielkrei Thank you 🤩

[@danielkrei]

Sorry for late response :) Here you go: ruclips.net/video/23k4okrrH7A/видео.html

[@world_production]

@@danielkrei Awesome ! Thank you 🤩

[@gileneusz]

Can you visualize 150B parameters? Thanks!

[@seer775]

bruh, here he has shown a few thousand. That is impossible with today's technology

[@chrismofer]

@@seer775 probably not impossible, would just take a long time to render out an image that somehow contains all of them.

[@TeuFortMan]

🤣🤣🤣

[@aeneaslee3652]

Hi there, I know a bit about manim,too. But manim can't process existing video files very well(That means you can't import a video in manim and make animation with that) So I am wondering that how do you do to add manim text to the video? Thx a lot! Respect:)

[@danielkrei]

I use DaVinci for editing. I export segments in Manim and use them in editor after

[@JonahNelson7]

Can we use this video in a video where we briefly mention neural networks? We need a nice visualizer. Thanks

[@danielkrei]

Sure! Just leave a link to this video in the description:) This helps a lot!

[@Blackkspot]

I didnt get the last part. what was happening in the convo network in detail? please 😊

[@danielkrei]

Sorry for the late response!
The convolutional neural network has only one convolutional layer with six filters in this case. So the input is turned into 6 feature maps, which are latter max-pooled (that is where the resolution of feature maps decrease).
After that, simple linear layers are used - flattening operation is not visualised.

[@flareonspotify]

cool

[@pocketgamer8446]

Hey, that was very interesting, I finally understood what a CNN is really... btw, with what software did you do this animation ?

[@danielkrei]

Hello, thank you!
I used blender for the animation. I made a video that explains the idea behind it. Check it out if you want to reproduce it: ruclips.net/video/23k4okrrH7A/видео.html

[@pocketgamer8446]

@@danielkrei Thanks !!! 🙂

[@philtoa334]

Very Nice.

[@Blooper1980]

No do a transformer.. Haha

[@handcrafted30]

Literally zero idea what I just watched