TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Quickdraw with Google AI

132 点作者 aloknnikhil大约 4 年前

26 条评论

gregsadetsky大约 4 年前
Many people here have pointed out that the data is available -- <a href="https:&#x2F;&#x2F;github.com&#x2F;googlecreativelab&#x2F;quickdraw-dataset" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;googlecreativelab&#x2F;quickdraw-dataset</a><p>I toyed with it some time ago and was impressed to see that -- in the raw data set -- each drawing is a time-series of strokes i.e. you get a person&#x27;s drawing (of a penguin, the Mona Lisa, etc.) after 0.1 second of drawing, after 0.2 seconds, etc.<p>I made and (shameless plug!) sell giant algorithmically generated posters which illustrate and play with this idea: each group of 8 rows shows doodles after a decreasing number of seconds. It&#x27;s fascinating to see how someone&#x27;s drawing of a &quot;koala&quot; is pretty much set for &quot;success&quot; after a few strokes, while other drawings don&#x27;t ever really improve with time... or how universal and &quot;well-defined&quot; are drawings of matchsticks, unlike those of kangaroos.<p><a href="https:&#x2F;&#x2F;gumroad.com&#x2F;l&#x2F;quickdrawposter" rel="nofollow">https:&#x2F;&#x2F;gumroad.com&#x2F;l&#x2F;quickdrawposter</a> if anyone&#x27;s ever so inclined!<p>Cheers
评论 #26630634 未加载
评论 #26630511 未加载
muglug大约 4 年前
Maybe needs a (2016) in the title.<p>Past discussion: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12965311" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12965311</a>
numlock86大约 4 年前
What is amazing takes the fun away at the same time: The detection is way too fast&#x2F;early. For example you are tasked to draw an octagon, the detection will already trigger once you drew three lines that are roughly aligned at 135° to each other. Draw an elephant? I can see there is something that remotely looks like a trunk and there is an attempt to draw an ear that&#x27;s merely a stroke at this point. It&#x27;s an elephant alright!<p>Things like that made me play the game in a different way: Draw it so the AI won&#x27;t detect it but a human would most likely instantly recognize it. Example: Microphone. Just draw a microphone hanging from a two-segment microphone arm. Start with the arm and add the microphone last and draw a circle around it or point at it. Clearly someone would recognize this as a microphone, while the AI struggles and thinks it&#x27;s something completely unrelated.
评论 #26632523 未加载
评论 #26632898 未加载
tyingq大约 4 年前
&quot;Draw Flip Flops in Under 20 Seconds&quot;.<p>And, of course, I wasted half of it trying to remember what a latch circuit looked like. Before remembering they probably meant beach shoes.
OnlyMortal大约 4 年前
I read the title as if it was some kind of classic Macintosh API to Google’s “AI”.<p>The title needs rethinking.
jdblair大约 4 年前
When I read the headline, I thought it was QuickDraw, and it was going to be some kind of classic MacOS &#x2F; Google mashup.
评论 #26632197 未加载
tonyteate大约 4 年前
As a few other other have pointed out, the data from this game is open-source and accessible in number of ways [1].<p>For an introductory exercise to deep learning for image classification, it&#x27;s a great alternative (or follow-up) to the classic MNIST dataset [2], which serves as a common &quot;Hello, world&quot; for image-based ML.<p>We trained and deployed such a model for a demo of our mobile ML tool [3][4]. Feel free to ping me if you&#x27;re interested, would love to chat.<p>[1] <a href="https:&#x2F;&#x2F;github.com&#x2F;googlecreativelab&#x2F;quickdraw-dataset#get-the-data" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;googlecreativelab&#x2F;quickdraw-dataset#get-t...</a><p>[2] <a href="https:&#x2F;&#x2F;codelabs.developers.google.com&#x2F;codelabs&#x2F;cloud-tensorflow-mnist&#x2F;#2" rel="nofollow">https:&#x2F;&#x2F;codelabs.developers.google.com&#x2F;codelabs&#x2F;cloud-tensor...</a><p>[3] <a href="https:&#x2F;&#x2F;imgur.com&#x2F;UagduZX" rel="nofollow">https:&#x2F;&#x2F;imgur.com&#x2F;UagduZX</a><p>[4] <a href="https:&#x2F;&#x2F;palletml.com" rel="nofollow">https:&#x2F;&#x2F;palletml.com</a>
Wowfunhappy大约 4 年前
Is there any way to get to a “free-play” mode? I’d like to just try drawing stuff and see if the AI gets it.
allwynpfr大约 4 年前
Link to previous discussion on HN, 4 years ago! <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12965311" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12965311</a>
chipgap98大约 4 年前
I assume this is picking from a predefined set of things? As in not all nouns, but a smaller list of possibilities. Otherwise it is pretty impressive that it went 6 for 6 with my terrible drawing abilities.
评论 #26629374 未加载
评论 #26630581 未加载
_Donny大约 4 年前
I drew a beautiful sea turtle; I was sure it was going to guess it! But no, it couldn&#x27;t. Then I checked the training set, and it was full of tortoises...<p>I guess they should check their training set :P
_joel大约 4 年前
Hrm, I&#x27;m not sure maybe it was a fluke, my drawing is terrible <a href="https:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;R8KE87K" rel="nofollow">https:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;R8KE87K</a>
评论 #26631272 未加载
rickdeveloper大约 4 年前
I made a camera version of this a few years ago (<a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=GEUrMbx_uac" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=GEUrMbx_uac</a>) using QuickDraw to train the model. I generated thousands of paper-like backgrounds (grayish noise) and pasted the raw drawings on top of it to generate JPEGs.
wdutch大约 4 年前
I love using this as an English Foreign Language teacher. I just wish I could choose a subset of the vocab because some of the words are a bit obscure like &quot;megaphone&quot;
JrProgrammer大约 4 年前
&quot;Draw the mona lisa in under 20 seconds&quot;<p>Who do you think I am?
ChrisArchitect大约 4 年前
(2016)<p><a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12965311" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12965311</a>
BugsJustFindMe大约 4 年前
Hmmm...<p><a href="https:&#x2F;&#x2F;m.imgur.com&#x2F;a&#x2F;y73R5Ir" rel="nofollow">https:&#x2F;&#x2F;m.imgur.com&#x2F;a&#x2F;y73R5Ir</a>
评论 #26632265 未加载
评论 #26629592 未加载
Google234大约 4 年前
Well, I learned the I’m bad at drawing :(
holistio大约 4 年前
It doesn&#x27;t recognise dicks, even though I&#x27;m pretty sure it&#x27;s, uhm, well trained for that.
poglet大约 4 年前
Draw: Sock &quot;I see socks&quot; &quot;sorry I couldn&#x27;t guess it&quot;
scaraffe大约 4 年前
A bit off-topic, but is there a dataset like this for memes?
aerodog大约 4 年前
Something tells me this would make for a good IQ test
dirtyid大约 4 年前
Recognized my poor squibbles better than my parents.
vekker大约 4 年前
Does anyone know if this is available as an API?
评论 #26629742 未加载
评论 #26629005 未加载
评论 #26629527 未加载
amelius大约 4 年前
Warning: emits audio.
domatic1大约 4 年前
Fun thing to do is draw a penis first and then start the real drawing