31–40 of 76 posts
Earlier quoted context omitted.
looking into it, I will update the results if I find an error!
Also, the 3rd chart in "Overall Comparison" has a #REF! label
Great comparison, and thanks for the review, happy being in the TOP-5 :). It seems, though, Sightengine's scores on the SFW category were swapped. The detailed google sheet lists the NSFW "raw" scores, not the "safe" scores.
Check out the "links to original images" sections as they are incorrect ~50% of the time. The name above the blurred image does not correspond to the link name.
I'm aware of the issue, and trying to fix it. But medium won't let me make any edits. Since I originally published the blog have been in constant touch with their support but they all seem to be away for thanksgiving. Just keep getting the error "Oops! Something happened to Medium. We’ll fix it as soon as we can."
I posted the blog here again with fixes https://www.adityaananthram.com/comparison-of-the-best-nsfw-...
The Yahoo NSFW classifier that is mentioned is available here: https://github.com/yahoo/open_nsfw I wonder if there's a way to ship this model and run it in an end-user app, in a way that does not require additional setup. I could use that.
Take one frame per second and check it using these services? But that would get super expensive real quick ..
Great write up. As a photographer, the more companies are relying on this the more frustrating the mis classification of SFW as NSFW is getting. I’ve copped a few Facebook bans from automated detection (I assume given how quickly the report comes in) for images totally within their community guidelines. A number of my photographer friends are in the same boat also. I counted at least 5 of us on 30 day bans in October…
I had a friend pick up a presumably automated ban from Facebook this month for posting a picture of a soldier from a WWII re-enactment he went to. No gore, no nudity, just a half body shot of a German soldier in uniform standing in the camp, smiling at the camera. No idea what triggered it, but it is scary.
A very interesting and related thing are images generated using the Yahoo NSFW classification neural network. It's fairly NSFW: https://open_nsfw.gitlab.io
This is brilliant, especially the cross-categorised images. I had thought these were talking about 'pornography', but this makes it clear that a very large amount of its training set must consist of dick pics.
This is very cool comparison. I would really like folks at Netflix/Amazon to take this and build a feature that auto-skips through non-PG rated content so you can watch some of otherwise very great movie/TV content with very young kids. For example, StarTrek. They can call this feature "PGify" :).
Just think, someone had to train these models. There are also models trained specifically for child pornography. Would you like to know these people?