As someone who just signed up for a computer science degree, this is highly interesting and well presented. Keep up the good work, youtube algorithm finally found something proper from a small channel!
@ProgrammingWithJulius
Ай бұрын
Thanks for the kind words, it means a lot! This is the first comment this channel has received by the way :) If I can give you some honest advice for your studies: It will be hard, but after the first few „filter“ courses like maths and so on, it will get better. During my degree I barely learned anything useful (for real programming work) since everything was highly theoretical. What helped me land a job after university was my years of practical experience in coding. So if you want to be successful, it’s best to pick up coding as a hobby and to just dish out some cool side projects :) Good luck!
@kibabyte
Ай бұрын
@@ProgrammingWithJulius Great to hear! I was already told the whole "filter" thing would be the case by friends, good to know that this is somewhat widely the case. I already do quite a few different things related to the field as a hobby. Thanks for the advice! Also a written, more in depth version of the videos you make would be greatly appreciated I think.
@ProgrammingWithJulius
Ай бұрын
@@kibabyte I had to leave out many details to this project, for example: - Ablation studies (training with different amounts of training material, different data augmentation strategies, network sizes etc.) - The many problems I had with the screenshots - Steam doesn't like it AT ALL when there are 150k images in the internal screenshot db. I basically had to nuke my whole Steam installation after I took them because I got like 10 fps and constant crashes in some games - Some other cool stuff like activation maps for the network (basically which patterns the nn looks for in a map) However, editing the video took WAY longer than programming for it. I was just done at some point :D When I get the time, maybe I will clean up and release the code, dataset and model.
@JooT471
Ай бұрын
Very well made Video - happy that the algorithm somehow sent me here
@JimRaid
Ай бұрын
Yo, great video, man! Keep 'em coming! 🎉 Just a heads up, the title was a bit confusing. I thought it was about predicting the next map in a casual game, not identifying a map. But I guess that's the technical term for it, huh?
@ProgrammingWithJulius
Ай бұрын
Thanks for the feedback dude :) I‘m afraid so, but I’ll pay more attention in the future when picking titles.
@EmilPrivate
Ай бұрын
Nice video, looking forward to your other projects!
@ProgrammingWithJulius
Ай бұрын
Already working on the next idea :)
@EmilPrivate
Ай бұрын
@@ProgrammingWithJulius Any spoilers? 👀😄
@ProgrammingWithJulius
Ай бұрын
@@EmilPrivate because of the positive feedback, the next one is going to be a follow up to this one, with some cool new features :) But I've got a huge list of ideas to work on :D
@EmilPrivate
Ай бұрын
@@ProgrammingWithJulius Exciting! Goodluck!
@skurtlil1104
Ай бұрын
Nice karamabit ( Commented to get another impression for the algorithm)
@ProgrammingWithJulius
Ай бұрын
It actually helped 😂
@Gansu360noscope
Ай бұрын
damn 82 subs? i expected 50k+ youre absolute sigma and madlad broski...
@ProgrammingWithJulius
Ай бұрын
Hoping to get there :)
@Chair991
Ай бұрын
Not too long ago I did an assignment in our machine learning course where we used neural networks, so it's actually really cool to be able to understand everything that's going on
@cookiecz123
Ай бұрын
Really nice
@tigerxplso
Ай бұрын
Very well made video. One tip, when you list numbers like in 1:20, please for god's sake, place them on screen, i had no idea where to look at, and could not remember any numbers you were listing, didn't know why you were giving an example, is it related to something later on etc.
@AjinkyaBhushan
Ай бұрын
great video. Idea: Ai to predict which side will win the match.
@ProgrammingWithJulius
Ай бұрын
Thanks for the idea, I’ll put it on my list :) I would need a large amount of demos for that though. Perhaps when I’m bigger I can ask leetify to give me some 😬
@Xapyion
Ай бұрын
Would love to see more vids like this
@shorty1276
Ай бұрын
Cooles video, Ich freu mich auf mehr :)
@tjorbenxx7549
Ай бұрын
very good vid! keep going
@ProgrammingWithJulius
Ай бұрын
Thank you!
@deltahl
Ай бұрын
3:20 If you have OBS installed you can enable the websocket server and use it to capture screenshots. (It's returned as a PNG encoded in a base64 string) Its faster and doesn't require focus on the game window.
@deltahl
Ай бұрын
I've written a script for it in python. (gist douglasb78)
@ProgrammingWithJulius
Ай бұрын
Thanks, this could come in handy in the future! I experimented with video capture as well, but since the images were changing so rapidly, there were many compression artifacts. Perhaps if I didn’t teleport randomly but moved smoothly, this approach would work better.
@somebodythatyouusedtoknow6191
Ай бұрын
Really great video
@ProgrammingWithJulius
Ай бұрын
Thanks!
@Ninja_Sandwich
Ай бұрын
Super cool video! Nice!
@ProgrammingWithJulius
Ай бұрын
Thanks!
@michaelsnickers3910
Ай бұрын
Great 3kliksphilip reference
@StiekemeHenk
Ай бұрын
One hit wonder haha, sad that not a lot of people subscribed. Maybe push that earlier and harder in the video?
@Proplom
Ай бұрын
nice
@wildxgamer7900
Ай бұрын
Can you try giving this AI the color palettes of the maps instead of an image of the map, and see if it can determine which map it is?
@ProgrammingWithJulius
Ай бұрын
How would you define color palette? I could for example take all the pixel colors of all the training data for each map, but I'm sure that almost any color is contained in almost every map. I'm not sure how I should quantify that dust2 is "browner" than inferno, for example. If you have any ideas, I could try it :)
@Astrobay13
Ай бұрын
Someone got tired of losing CS-guessr :D Could it be trained to pinpoint the position on the map?
@ProgrammingWithJulius
Ай бұрын
Great addition to my follow-up ideas for this project, thanks! Since I’m setting the location via console, I always know where I am. I’d just need to retake the screenshots to save the location as well. The model should be able to learn that.
@wehushi_sushi
Ай бұрын
What is you show it some pictures from csgo instead of cs2 of the same maps?
@ProgrammingWithJulius
Ай бұрын
Sadly I can’t run a quantitative analysis, but I did try giving it some screenshots from the internet. The result: it depends. Newer maps like nuke have a shoddy success rate, but maps like dust2 or mirage work well.
@le3lux
Ай бұрын
That's awesome! Can you somehow release a tool with the AI so anyone can use it? That would be awesome!
@ProgrammingWithJulius
Ай бұрын
I am working on a version that’s releasable. I’m also preparing a dataset for a public release - stay tuned! Can you tell me what your plans are with the model? :)
@monkeytime3171
Ай бұрын
only 60 subs?
@ProgrammingWithJulius
Ай бұрын
Just started the channel, a couple of days ago I had 0 :)
Пікірлер: 47