TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

What a Deep Neural Network thinks about selfies

262 pointsby vkhucover 9 years ago

14 comments

lqdc13over 9 years ago
A guide on how to take a good selfie that others will like:<p><pre><code> be female be blonde be attractive </code></pre> Incidentally, Christian Rudder did a really good &quot;study&quot; on the dating site pictures a few years ago:<p><a href="http:&#x2F;&#x2F;blog.okcupid.com&#x2F;index.php&#x2F;dont-be-ugly-by-accident&#x2F;" rel="nofollow">http:&#x2F;&#x2F;blog.okcupid.com&#x2F;index.php&#x2F;dont-be-ugly-by-accident&#x2F;</a>
评论 #10449050 未加载
评论 #10449350 未加载
评论 #10452747 未加载
评论 #10450574 未加载
评论 #10450256 未加载
danblickover 9 years ago
This is neat. I bet Facebook or OkCupid are sitting on all sorts of click data that could be used to develop tools for helping people make their photos look better. (Even if, personally, I can&#x27;t wait for a cultural backlash against internet narcissism...)<p>[Edit: Even better, he didn&#x27;t use click data to train the model, just public likes.]
评论 #10450106 未加载
anunderachieverover 9 years ago
I would like to see a deep dream selfie ...<p>Feed it an initial picture (noise, clouds, a selfie) and then backwards manipulate the input to maximize the assessed quality of the &quot;selfie&quot;.<p>I guess that would look pretty funny.
评论 #10449250 未加载
misiti3780over 9 years ago
One thing I always found interesting is Lecun is credited with developing covnets, but Hinton is apparently credited with scaling them and showing the world how great they are in the paper from 2012 - why was Hinton&#x27;s group (Toronto) able to publish these ground breaking results before Lecun&#x27;s group (NYU)
评论 #10449065 未加载
评论 #10449071 未加载
nightpoolover 9 years ago
&gt;Be female. Women are consistently ranked higher than men. In particular, notice that there is not a single guy in the top 100.<p>This sounds true, but it can&#x27;t be the real reason—selfies are ranked relative to the other images by the <i>same user</i>. So unless users are taking a lot of #selfies of people of different genders, we can assume the dataset is already controlled for the gender of the person in the image, no? Unless there&#x27;s some confounding factor at play, such as some demographic segment being more likely to optimize for good selfies occasionally but have boring feeds the rest of the time.<p>would be super interesting, if the data is available, to normalize this by exposure. Of the people that saw an image, how many clicked &quot;like&quot;?
评论 #10449339 未加载
评论 #10449142 未加载
JabavuAdamsover 9 years ago
How to take a good selfie: don&#x27;t be black or dark-skinned, unless you&#x27;re a celebrity.<p>How do we prevent our AIs from learning racism?<p>EDIT&gt; Informative article, BTW. A good read.
评论 #10451835 未加载
评论 #10449811 未加载
vonnikover 9 years ago
I think it&#x27;s less about the head getting chopped than about having &quot;the head take up about 1&#x2F;3 of the image,&quot; as Karpathy says. So what the net is learning is composition, or balance in an image, which is really cool. The rule of thirds is actually pretty well know to people in photography:<p><a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Rule_of_thirds" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Rule_of_thirds</a><p>(Our deep-learning framework <a href="http:&#x2F;&#x2F;deeplearning4j.org" rel="nofollow">http:&#x2F;&#x2F;deeplearning4j.org</a> missed his list, but it&#x27;s got working convnets, too.)
评论 #10449207 未加载
netheril96over 9 years ago
One caveat with these machine inspired knowledge: they are prone to error, probably more than humans, at least for now.<p>For example, if you train a CNN directly with human faces, its recognition rate comes way below what a human is capable of. Only after you apply tons of handcrafted optimizations, which are mostly black art, will you get close to or surpass a human&#x27;s capability. Without much domain specific tuning, an AI&#x27;s insight is far from reliable.
评论 #10449822 未加载
评论 #10449367 未加载
RealityVoidover 9 years ago
It seems this neural network has a sense of humor if you look at the &quot;Finding the Optimal Crop for a selfie&quot; area.<p>You can see it optimized the last selfie by cropping the face fully out of the picture.. :))
spikelsover 9 years ago
DNN is a key technology of the future. I highly recommend the education program Professor Karpathy mentions at the end of this post. All are excellent and free.
amaiover 9 years ago
I have seen similar results before: <a href="https:&#x2F;&#x2F;medium.com&#x2F;the-physics-arxiv-blog&#x2F;the-algorithm-that-sees-beauty-in-photographic-portraits-435ab8064646" rel="nofollow">https:&#x2F;&#x2F;medium.com&#x2F;the-physics-arxiv-blog&#x2F;the-algorithm-that...</a>
JoachimSover 9 years ago
A really good read. Good intro to ConvNets, a well designed and implemented test. Ad funny.
trhwayover 9 years ago
looking at the top 100 one can only wonder how Hollywood has figured it out well before mighty power of computer :)
评论 #10449039 未加载
thewhitetulipover 9 years ago
Well, you don&#x27;t need to ask a deep neural network to say that selfies are getting stupid daily with teens sticking their tongues out
评论 #10450155 未加载