Why Quick, Draw! Is Sharing Millions Of Doodles With Everyone

Why Quick, Draw! Is Sharing Millions Of Doodles With Everyone


NAT: Doodle alert. Doodle alert. 50 million doodles alert. [MUSIC PLAYING] Hey there, I am Nat,
and today I have a quick announcement for you
about the game “Quick, Draw!” If you haven’t played
it yet, “Quick, Draw!” is a game where in
20 seconds or less, you are asked to draw something,
and then a neural network tries to guess what it
is that you’ve drawn. QUICK DRAW : Oh, I know. It’s moon. NAT: Since it
launched last year, more than 15 million people
from all over the world have played it and drawn
millions and millions of different bicycles, and
couches, and chandeliers, and crocodiles. There are over 300
different objects that the game asks people to
draw, resulting in millions and millions of drawings. Now the team behind the game,
including several people that I work with, have decided
to release a large data set of these doodles
to the public so that people can
learn from them, and make new things with
them, and just admire this gigantic collection of
art made by people everywhere. How you can see these drawings
is if you go to the “Quick, Draw!” website,
you’ll now see a link that says, “the world’s
largest doodling data set.” Click that, and you’ll
get to the gallery of all these drawings and
have the option to download all this data. I’ve had an early peek
at these drawings, and I find them
fascinating and almost mesmerizing to scroll
through and just observe all the
different variations on the same common object. Like for cat, some people
draw just the cat head. Other people draw
the whole cat body with a tail and everything. For camels, some people
like to draw camels with one hump versus
two versus three. I want to make a
future video that dives deeper into what we can
learn from studying this data and what new research
it’s helping to enable. And to make this video,
I need your help. So first, if I talk with
anybody associated with the game “Quick, Draw!” people that are doing new
research with this data, what do you want to know about? Leave me a comment on this
video with your questions. And secondly, now that the
data is out and available and you can take a look at it,
are there any observations, any patterns you see? Let me know what you
think is interesting, and if you have any explanations
for these doodle phenomena. You can also leave a
comment on this video, or I now have an
email address where you can send me your thoughts. That’s all for today. So thanks for watching. Hope to hear from you soon. And just in case you
didn’t get the memo. Doodle data alert. Double data alert. Doodle data alert. Bye. [MUSIC PLAYING]

100 thoughts on “Why Quick, Draw! Is Sharing Millions Of Doodles With Everyone

  1. People should draw different objects for the mentioned names just to fuck with the computer. Ex: draw an ice-cream when asked about a computer.

  2. Hey Nat!! I would love too see a unified vector of all of this draws. like just one cat draw that is the combination of most of well drawer cats. it would create a new icon that would represent how people see things of the real world in the actual 2d screen or in a paper. See, it would be the most accurate version of some real world object transformed into a icon. I thinking to do that on my on cause that really excites me!!

  3. Are they planning on adjusting for the fact that some drawings have extra detail? Like the fact that 'beard' drawings frequently have a face too, despite that the entire face is not a beard.

  4. as we can see many people draw different types of figures for the same thing…then how does the AI recognize them to be one?
    how are the data sets stored?

  5. I really wanted to compare bear and teddy bear and try and figure out why the neural net guessed bear for my drawing but not teddy bear (which is what I was trying to draw lol). Sadly, the teddy bear data seems to be missing πŸ™

  6. >was told to draw a rake
    >didnt know what a rake was
    >* googles rake *
    >scary images pop up
    >brain insta imprints images permanently in memory
    >nightmares for ever

  7. The NN is simply awesome! Question: How does it know when the classification is perfect, since it seems to stop attempting after getting the right answer?

  8. 1. How does AI recognize false-positives? People drawing a sun while AI want's to learn how a bike is drawn or so. Is the AI the way connected that it can alert a drawer that camels have two hops while he's drawing it?
    2. That's the task of the AI I think. So finding the patterns and recognizing it is something we don't have to explain anymore.

  9. Hey NAT, I'm a new follower of your channel and I just wanted to let you know that I'm really glad you created this channel! I've learned so much! It's so hard to find good female youtubers that make great vids. Keep up the great work! PS, u pertty πŸ˜€

  10. Does having the game only in english and the internet reduce the size of the sample of data collected? I assume most of the data collected comes from people from the Y or Z generation speaking or understanding english and able to afford a computer. Does the team working on that has ideas to collect data from others targeted audiences? Could this data could used in some way to get better "real" image recognition software?

    PS: Nat I really love your videos, really interesting, showing the parts of Google we don't know about and that their is a lot of people behind the things we use every day.

  11. It would be interesting to see a model capable of making doodles of things it hasn't seen a doodle of yet (e.g. by combining the doodle data with pictures that are not doodles).

  12. I'd love to see how different cultures and ages influence the doodles! It would be really interesting to find other things that might also influence the doodles! gender? education level?

  13. Hey Nat,

    what do you work at Google on? πŸ˜€

    And question for the Quick Draw Team:
    How's Convolution Neural Network of Quick Draw build?
    What Layers does the network have and so on πŸ™‚

  14. Question for the researchers: what's the benefit of teaching a neural network to recognize drawings, as opposed to teaching it to recognize photographs of the same objects? Wouldn't the photos give the network a better idea of what the world looks like?

  15. quick draw is super fun, but i wish it were available as an android app in the play store. luv, a doodler.

  16. the randomize button on the website randomizes what picture you're looking at but not what order they were drawn in. seems like the first 1000 or so drawings are always going to be the same first bunch of people on the system rather than a random selection of entrants.

  17. Is it possible to guess the drawing before the user completely draws it, like GBoard guesses, by just binding the mean time from the data set that users took to draw a single category? It would definitely be pro-active suggested drawing. TensorFLow is Magic though!

  18. If you feed it with a random input, and tell it to relax down the steepest descent to be more like a cat, does it produce convincing looking cat sketches?

  19. it's really surprising how for over 100 years a bicycle shape consisted primarily of 2 circles and 2 triangles yet virtually no one draws this. this is especially surprising given how common bicycles are around the globe.

  20. My question: How does the game recognize certain objects if they are all drawn so differently? Does it break the drawings up into different shapes? How does it work?

  21. 1. Do you think the public is strong willed enough to get the program to believe an object is something else by drawing a different object or is the majority of individuals overwhelmingly good and they only draw the object listed?
    2. Will the program ever be smart enough to guess right the first time no matter who is drawing?

  22. Hello, Nat, i is Kordell, it is difficult to express myself since I have dyslexia. Even though I have a creative mind. I never write down my idea Since I would stumble In the first few words. or for fear of my mistakes making me Nervous about what I was writing in the first place Currently, I use Google Translate Along with Grammarly Just great for comments sections like this, however, I'm still unable to be a part of a live stream or Other social websites Google translate doesn't always get the words right And it takes far too long To get my entire idea are crossed With a simple image or quick drawing Far less headache inducing end A lot easier for me to understand. I would love to see this translate words damages and vice a verse This comment alone has taken me 20 minutes or so to write.

    Β I would love to know what you think Sincerely Kordell

  23. pretty neat way of crowd sourcing the learning curve for a computer to learn things. I suppose language could be learned in a similar way?

  24. Just downloaded the dataset from GitHub. I now have a lot of interesting ideas for projects that I could do involving this data. Thanks for letting us know!

  25. How does a machine learning AI extract and "average" drawing from this data if certain doodles can be drawn two different ways, like 50/50? For example, If the AI was asked to make an average cat doodle, would it make one with a body, or just the head? Or would it be a creepy mix of the two? and why would it choose this?

  26. What inspired them to start this project? Did they have this idea of sharing it right from the start? What are they expecting now that it is out there?

  27. It can be cool if one day a year there will be on Google page doodle created from all this data like jitter style. If that's possible…

  28. Is Google using these Drawings to integrate them in some AI app that can recognize the thing we draw?

  29. unfortunately this feature isn't available in Iran and none of us can use it and always we see the 403 error. IDK why google block these useful features for us.

  30. Unfortunately we can't use lots of Google services in Iran or in another hand, other services that provided by US. IDK why they set these ridiculous sanctions to us.

  31. would be nice to see where people stated on the doodle and what order they draw things on that data sheet

  32. Does Google actually use Quick Draw game for any internal purpose? How about doing a video on the Crowdsource app next?

  33. I love google and your way of explanation thats make me think differently and popup with a lot of ideas please keep it up and try to do 2 videos per week and go more deeper in google with great place to work and create innovation thats help WORLD πŸ™‚

  34. sometimes people draw nonsensical zigzags when they sort of give up on the task to draw complicated items … like me lol
    does Google attempt to remove such "seemingly gibberish" images from the database?

  35. Thanks for sharing all this knowledge with the Internet! This is by far one of my favourite channels. Here's my question:
    At the website they mention that the learning process uses not only the drawing itself, but also the order of each stroke. I'm curious on how do they incorporate this order as input for the neural network, or if this is a completely separate learning process.
    Congrats on your work both at Google and here! πŸ˜€

  36. you could order the drawings of a specific item by complexity. print it out a mile long. Or maybe you could find their closest real world matches.

  37. When you scroll trough an object in the quick draw data page, you see all the diffrent doodles, but you realise that as much as there are of theese doodles, none of them is the same. That gave me the chills.

  38. salam je ss aziz aarich homminal bleu servical .rien qui s appel doddols ce sont des desins des marcher et des boursses de tous les marcheΓ©s mottons vaux poulet bleΓ© riz argent alle markettes in ternational parceque ils ont fini avec becam adodols .derigeΓ© par les jenne bleu qui leur die les prix avent tous le monde des divers marcheΓ©s

  39. πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜ πŸ˜ πŸ˜ πŸ˜ πŸ˜ πŸ˜ πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜‡πŸ˜‡πŸ˜‡πŸ˜‡πŸ˜‡πŸ˜‡πŸ˜‡πŸ˜¬πŸ˜¬πŸ˜¬πŸ˜¬πŸ˜¬πŸ˜¬πŸ˜ˆπŸ˜ˆπŸ˜ˆπŸ˜ˆπŸ˜ˆπŸ˜…πŸ˜…πŸ˜…πŸ˜…πŸ˜‘πŸ˜…πŸ˜‘πŸ˜…πŸ˜…πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜‘πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜‰πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜―πŸ˜πŸ˜πŸ˜πŸ˜²πŸ˜²πŸ˜²πŸ˜²πŸ˜²πŸ˜²πŸ˜²πŸ˜²πŸ˜²πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜πŸ˜žπŸ˜žπŸ˜žπŸ˜žπŸ˜žπŸ˜žπŸ˜žπŸ˜ž 😷😷😷😷😷😷😷😷😷😷😷

Leave a Reply

Your email address will not be published. Required fields are marked *