Thread by @mrityunjay_99, So the generative deep learning study session is about to end which [...]

Mrityunjay

mrityunjay_99

So the generative deep learning study session is about to end which means it's a good time to share some of the experiments I have worked on while reading this book along with amazing people @__MLT__

A generatively adventurous thread

1) trained a simple Autoencoder on MNIST ( classic example) + created some visualizer to walk along the latent space using @tensorflowjs owJS and @d3:-

2) trained VAE on some of the CG characters I've built previously ( still not satisfied with the result though...but oh well :P )

3) Style Transfer: using pretrained model from amazing peeps @magenta and instead of giving it a. Style image I though why not use the results from fluid sim and I have to say it looks pretty awesome

Some Stills:

4) Music: so, this one scared me for a while because of the fact that you have to train a really big network ( music RNN, museVAE etc) on your custom dataset to feel natural and since I have a potato laptop I couldn't afford to do that...

Enter the amazing people @GoogleMagenta and there MuseVAE and MidiMe which uses the concept of latent constraints which lets you get really plausible result with really small datasets and training time ( mind blown!) ...

So using the midiMe I was able to train it on some of the jazz battle,piano battles I like and boy do I love the results

... Please have a go at it, and let me know what you guys think...

5) some more music! Trained on deathNote soundtrack and it does capture that mysterious vibe! ....
https://soundcloud.com/mrityunjay-bhardwaj-411805942/ai_3-mysterious-loop

Also, for more generative soundtracks, I have created a little playlist of some of the results I liked! :- https://soundcloud.com/mrityunjay-bhardwaj-411805942/sets/ai-music

6) Play : So, for me this was the most interesting chapter of them all and as much as i would love to, i don't have anything concrete to share currently, still dabbling in TFJS version of world model .... i'll be sure to update this thread as soon as i have something concrete :D

Lastly, I would like to thank the super super awesome and inspiring @__MLT__ community because of you guys and because of this study group I get to learn soo many amazing things! + i get to blend both sides of my brain like never before!

Thank you so much

Also, Special thanks to our @anugrahasinha grah , @suzatweet @vatai @JsonMathsAi + thanks for some really inspiring experiment @girlandkat @shyam_shinde_ and @cataluna84 you guys rock!

That's it from my side!

Stay safe and have a beautiful day....

You can follow @mrityunjay_99.

Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: