Ask HN: What are the most fun-to-play-with open source deep learning projects? | Hacker News Reader

Ask HN: What are the most fun-to-play-with open source deep learning projects?

104 pointsjonplackett23 comments

I have temporary access to a monster PC (32core thread ripper + 2x2080ti).

I've been playing with deep fakes which is fun (I look good as Neo fighting Morpheus).

But what other fun + awesome open source projects can I run with it before I have to give it back?

I'm a programmer (making apps, websites, games etc). But don't have a lot of deep learning experience though want to learn.

Any suggestions?

Thanks!

View on HN

I've been playing with reinforcement learning on simulated robots, using OpenAI Gym[0] and PyBullet[1].

I find it interesting because, in theory, the learned policies transfer to real-world robots, and the time is nigh for development in that area. Plus you get to watch it do funky robot shit and slowly get better.

If you want a quick way to dive in, I have a repo[2] I've been using to train on a variety of PCs and Google Colab[3]. The task is a Panda robot pushing a randomly placed object. I had some trouble getting it all working with recent Keras at first, so my repo might save you time.

Email me[4] if you get stuck.

[0] https://gym.openai.com/

[1] https://pybullet.org/

[2] https://github.com/tlack/rl-experiments

[3] https://colab.research.google.com/drive/1HGmKMGwW_emok157ENr...

[4] lackner@gmail.com

Finetuning a GPT-2 model on logs from a chatting server like IRC or Discord you're active in, then writing a bot to generate messages from a prompt using the model.

Great idea thanks! Are there any other Generative models that are (relatively) easy to use? I find generative AI the most interesting.

Any good tutorials on this you recommend ?

I recently played around with a lib called Detectio which was super fun and met my needs.

Like you, I have barely any ML experience so it was nice to be able to still finish what I set out to achieve.

I write a simple tutorial and sample code base -> https://ajaxdavis.com/post/An-Open-Source-Computer-vision-mo...

Nice writeup!

I should of included a link directly to the repo README -> https://github.com/australia/aboriginal-flag-cv-model

LOL, that's not a monster PC for deep learning :) I was going to suggest https://github.com/openai/jukebox/ but with 2080Ti it's going to be tough. You might be able to do some sampling with the smallest lyrics set, idk.

Guess I need to do some deep learning learning...

Missed opportunity: deep Deep Learning learning...

Sounds like they need to do some deep deep deep learning learning learning.

it may not be a beast to you guys, but it's a pretty big step up from my Macbook Pro 2013 + GeForce GT 750M 2 GB

If you like deepfakes, you'll love playing with Runway ;) (based on StyleGAN transfer)

https://runwayml.com/

Or for a real competition, the OpenAI ProcGen Benchmark is really heating up!

https://www.aicrowd.com/challenges/neurips-2020-procgen-comp...

Have you done the https://www.fast.ai/ course? Paperspace also offers free Juypter notebooks once you need to return your computer.

Can I ask how/where you got access to that PC? Purely curious :-)

Through work. It's not needed currently, but I'm presuming someone will want it again soon!

Starlink satellites for microgrids ML. Classify power grids ammendable to MICROGRIDS ML. Use Geni.org power grid maps or California ISO maps. What is the right size for redundant wind/solar based microgrids?

I recommend Skoods (https://www.skoods.org/)

How do you generate deep fakes? I'd like to play around with that.

https://github.com/deepfakes/faceswap

or https://github.com/iperov/DeepFaceLab

which of those two are easier to get started with?

Face Swap has a graphical interface and the other is command line, but it has .bat files for you to just click and run so both are pretty easy to start with.

The main work is preparing the training data. Making sure the faces are detected correctly and not accidentally one of another person. It is very tedious work!

Reply on news.ycombinator.com