A deep learning game I've been working on that uses the awesome Three.js

This is a game I’ve been working on similar to 20Q, Akinator etc except that it is designed to work for any object, not just people etc. Anyway, I had a lot of fun building the UI using Three.js which has led me onto lots of other cool things. Thanks Mr.doob for creating such an incredibly awesome piece of kit. You rule.

You can check out the game at https://incredicat.com

It uses decision trees at the moment for the algorithm but I’m in the middle of updating it to use neural nets. Any feedback on the graphics (or game) much appreciated.

13 Likes

Pretty impressive! I was thinking “sandwich” and it gave me “hamburger” in 19 questions.
Second time was even more impressive. It guessed “screw” in only 15 questions!

2 Likes

Ha! Nice one. It’s still WIP but I’m hoping the transition to neural nets might unlock some further goodness…

Thought “crystal” and guessed “diamond.” Pretty impressive! The presentation’s great too.

Really cute! I love this cat :heart_eyes_cat:

I was thinking of a chess piece :chess_pawn: but it could only guess a baseball bat :cricket_bat_and_ball: :thinking:

Pretty impressive! I was thinking “sandwich” and it gave me “hamburger” in 19 questions.

@marquizzo I think the cat simply suggested an improvement :rofl:

1 Like

after some 30 questions it failed to guess wrist watch, but it had the phrase in the submission suggestions list. why?
Screen Shot 2020-03-05 at 19.07.06

oh boy, is it dumb, sorry. after asking “is it found in the sky” twice per single session it proceeds to ask me “is it found in the workplace/office” or “does it have corners” (what ?). and no, cat, it was a spaceship, not a helicopter.

1 Like

I also tried spaceship, it gave me space shuttle :slightly_smiling_face:

thats close enough.