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.

Open Sourcing a Deep Learning Solution for Detecting NSFW Images

453 pointsby pumpikanoover 8 years ago

36 comments

brianwawokover 8 years ago
So can it be reversed to become the ultimate porn-finding neural network?
评论 #12614544 未加载
评论 #12614630 未加载
评论 #12614487 未加载
评论 #12614662 未加载
评论 #12614507 未加载
评论 #12615336 未加载
评论 #12615078 未加载
评论 #12615514 未加载
评论 #12616070 未加载
bahroover 8 years ago
I should update my sexy map finder: <a href="http:&#x2F;&#x2F;exclav.es&#x2F;2016&#x2F;05&#x2F;20&#x2F;sexy-maps&#x2F;" rel="nofollow">http:&#x2F;&#x2F;exclav.es&#x2F;2016&#x2F;05&#x2F;20&#x2F;sexy-maps&#x2F;</a>
评论 #12616056 未加载
评论 #12616458 未加载
inlinedover 8 years ago
Forgive my ignorance of ML but the last bit: &quot;you&#x27;ll need your own porn to train on&quot; confused me. Does this mean that they&#x27;re just exposing the rough topology of their neutral net (eg depth) and not the actual weights between nodes? I&#x27;m curious to learn from an ML expert how much this actually offers.
评论 #12614566 未加载
评论 #12615090 未加载
评论 #12616253 未加载
wildpeaksover 8 years ago
Direct link to Github: <a href="https:&#x2F;&#x2F;github.com&#x2F;yahoo&#x2F;open_nsfw" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;yahoo&#x2F;open_nsfw</a>
zfedoranover 8 years ago
Has anyone tried taking the features that are learned at the various layers of a neural net and feeding them into something like this: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12612246" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=12612246</a>?<p>I imagine we would get some really interesting images back...
echelonover 8 years ago
Can you run deep dream on this? That would be quite fascinating.
评论 #12614708 未加载
NicoJuicyover 8 years ago
We are not releasing the training images or other details due to the nature of the data, but instead we open source the output model which can be used for classification by a developer.<p>I&#x27;m guessing the one who had to input the data&#x2F;images had a fun time at work :p
评论 #12614631 未加载
评论 #12615710 未加载
darklajidover 8 years ago
They acknowledge that NSFW (or pornographic) is hard to define, a la &#x27;I recognize it if I see it&#x27;.<p>But looking at the meager 3 sample images I&#x27;m confused about the scoring already. Why is the one in the middle scoring the highest?<p>The question is an honest one. The two rightmost images seem to be interchangeable to me and are ~boring~: People at the beach. Is this network therefor already trained to include the biases of the creators?
评论 #12615695 未加载
eggoaover 8 years ago
I wonder if anyone at Yahoo tried using this to &quot;deconvolute&quot; noise into Cronenberg nightmare porn?
TrueDualityover 8 years ago
Sit back grab some popcorn. Lets see how long it takes people to start running data backwards to get new original porn.
评论 #12614940 未加载
评论 #12614627 未加载
lifeisstillgoodover 8 years ago
My first thought was from years ago, when I was pitching open source forensic services to London police (did not get far, bad Salesman that I am)<p>Cataloging, categorising pornography seized is a nasty job and one that cops across the planet might do better with good common OSS tools.<p>Hopefully this will help
c3534lover 8 years ago
My first thought: would probably be very useful for sites to crack down on inappropriate content.<p>My second thought: I could probably use this to find porn in unexpected places via a webscraping Python program.
m-i-lover 8 years ago
Good to see they&#x27;ve automated this (beyond the initial classification of training data). In the early days of the web, such filters were typically based on manually maintained lists of sites. I actually met someone at a party once whose full-time job was to surf for porn, to maintain the filter for a provider of IT services to schools (he worked for a company now called RM Education). He said it was his ideal job for the first few days, but soon grew tiresome (note that back in those days there wasn&#x27;t really any extremely objectionable material on the web).
SloopJonover 8 years ago
Anyone else see the irony in acknowledging that NFSW is subjective and contextual, but assuming that pornographic images are not?
评论 #12615285 未加载
评论 #12615321 未加载
joshmnover 8 years ago
I&#x27;m not a deep learning person whatsoever, but I do have an interesting use case that I won&#x27;t disclose publicly: Is there a way to build this, and output detections based on the, ugh, object it has detected?<p>e.g.<p>penis 0.94<p>vagina 0.01
评论 #12614978 未加载
评论 #12616914 未加载
slowmovintargetover 8 years ago
I think I&#x27;ll pass on browsing the the deep dream visualizations for this.
zuzunover 8 years ago
With access to Flickr and Tumblr it must have been very easy to create a huge training set for such a task.
prirunover 8 years ago
Aren&#x27;t there more important problems to work on than worrying about someone looking at naked people? This is just what we need: more effort spent on censoring and controlling people.
评论 #12618199 未加载
cvwrightover 8 years ago
Reminds me of this post from hackerfactor where he describes his own porn filter based on pHash.<p><a href="http:&#x2F;&#x2F;www.hackerfactor.com&#x2F;blog&#x2F;index.php?&#x2F;archives&#x2F;529-Kind-of-Like-That.html" rel="nofollow">http:&#x2F;&#x2F;www.hackerfactor.com&#x2F;blog&#x2F;index.php?&#x2F;archives&#x2F;529-Kin...</a><p>It&#x27;d be interesting to see a direct comparison of the two. Off the cuff, I&#x27;d expect the deep neural network to be more accurate and better at generalizing, but much more expensive to train.
Dim25over 8 years ago
another work in this field: &quot;Adult video content detection using Machine Learning Techniques&quot; PDF: <a href="http:&#x2F;&#x2F;colorlab.no&#x2F;content&#x2F;download&#x2F;37238&#x2F;470343&#x2F;file&#x2F;VictorTorres_MasterThesis.pdf" rel="nofollow">http:&#x2F;&#x2F;colorlab.no&#x2F;content&#x2F;download&#x2F;37238&#x2F;470343&#x2F;file&#x2F;Victor...</a>
johnnyoover 8 years ago
I&#x27;ll bet this would be a good tool for sysadmins or network administrators to run against their network and see what it finds.
chadsciraover 8 years ago
Awesome!<p>I have been using nude.js to do this ( <a href="http:&#x2F;&#x2F;s.codepen.io&#x2F;icodeforlove&#x2F;debug&#x2F;gMrEKV" rel="nofollow">http:&#x2F;&#x2F;s.codepen.io&#x2F;icodeforlove&#x2F;debug&#x2F;gMrEKV</a> ), which is hit or miss.
ganwarover 8 years ago
To be precise they are only releasing the already trained model. The associated dataset is not being made public.<p>Thus, it is meant to be for off the shelf use rather than being able to tinker with the network to produce nuanced results.
评论 #12615926 未加载
patrickaljordover 8 years ago
I wonder what would happen if we stopped firing people for watching NSFW images. I mean bosses look at NSFW images all the time and it sounds like a shallow reason to fire someone.
评论 #12617341 未加载
Joofover 8 years ago
Are there any other fairly basic image recognition problems that people want? I&#x27;d be happy to provide as long as a dataset is easy to collect.
CompanionCuubeover 8 years ago
Has anyone run this NN on the censored Facebook image?
askewover 8 years ago
Interesting that the photo of two women on the beach is given a higher NSFW rating than the photo of a man on the beach.
Happpyover 8 years ago
Could this work on mobile to detect 18+ content in images or video? Or would it be a trained library of 50mb+?
KennyCasonover 8 years ago
I literally just started working on this problem 2 hours ago &gt;_&lt;
zwindlover 8 years ago
That&#x27;s it! That&#x27;s what I&#x27;m looking for.
matheweisover 8 years ago
is this just the network or is it a fully trained model? The TechCrunch article suggests the former but the yahoo post the latter...
评论 #12616463 未加载
cftover 8 years ago
I hope this is ported to TensorFlow soon!
rasz_plover 8 years ago
oh silly Americans, its just tits
ykover 8 years ago
I would suggest, that the link should go to Yahoo&#x27;s blog post<p><a href="https:&#x2F;&#x2F;yahooeng.tumblr.com&#x2F;post&#x2F;151148689421&#x2F;open-sourcing-a-deep-learning-solution-for" rel="nofollow">https:&#x2F;&#x2F;yahooeng.tumblr.com&#x2F;post&#x2F;151148689421&#x2F;open-sourcing-...</a><p>which contains some technical details. (And furthermore, I guess the HN crowd has enough Internet experience to come up with stupid jokes of their own design.)
BinaryIdiotover 8 years ago
The Yahoo blog[1] post is far more interesting than this techcrunch &quot;article&quot;. Suggest changing URL to the Yahoo Blog please.<p>[1] <a href="https:&#x2F;&#x2F;yahooeng.tumblr.com&#x2F;post&#x2F;151148689421&#x2F;open-sourcing-a-deep-learning-solution-for" rel="nofollow">https:&#x2F;&#x2F;yahooeng.tumblr.com&#x2F;post&#x2F;151148689421&#x2F;open-sourcing-...</a>
评论 #12614856 未加载
Budover 8 years ago
So this is what Yahoo was up to for the last 10 years, instead of building any sort of security, keeping Yahoo Messenger working properly, or anything else of value? Heckuva job, Yahoo.