Resurrect safe mode / filter out people

+1 to this, this is my main concern at work

+1

Thanks Peter, this is great!

Please test and report impressions, it is already deployed to ppa:variety/daily (https://code.launchpad.net/~variety/+archive/ubuntu/daily).

Though I don't expect it to perform 100% well until it is released to everybody and more people start reporting, and even then it won't be 100%. The backend will also have to get automated NSFW-detection "logic" with time. This is not yet ready.

I added the daily ppa maybe a week after you posted it, turned on regular
wallpaper changes on both my work and home machines, and haven't yet had to
actually use the "Report NSFW image" option. I guess that means it's
working?

On Sun, Aug 7, 2016 at 10:20 AM Peter Levi <email address hidden>
wrote:

> ** Changed in: variety
> Status: Fix Committed => Fix Released
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1582885
>
> Title:
> Feature request: filter out people
>
> Status in Variety:
> Fix Released
>
> Bug description:
> Wallpapers with sexy ladies are not appropriate at work, but I would
> still like to use Variety. It would be useful if we could filter out
> images with people in them. This is obviously a very serious and
> important application for Google's deep learning results for image
> description.
>
> Alternately, I can just scroll through the selector manually and
> delete those I don't want to see.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/variety/+bug/1582885/+subscriptions
>

I wanted to chime in again with a question or possible bug about this
feature--I've used it a couple times to give ratings, but, almost as often,
I've opened the menu to give a rating and had no option to give one.

E.g., I got this image
<https://unsplash.com/photos/2QcARuHXUUw?utm_source=Variety+Wallpaper+Changer&utm_medium=referral>
this
morning, but only these options in the Image menu:
[image: menu.png]

In fact, in recent versions (I'm now on 0.6.8), I actually can't find any
mention of this feature in the preferences. Was it removed?

@edgarswooth I believe in newer versions of variety Peter has mentioned he's removing the ability to rate images: http://peterlevi.com/variety/news

Basically as I understand it he's proposing reducing the number of wallpaper sources to more reputable sites like Unsplash - making "safe mode" unnecessary/redundant.

Personally myself I'd rather not see a reduction in the number of wallpaper sources. But at the same time, I share the same concerns that the OP does about inappropriate imagery popping up especially on work computers.

My "solution" to the issue is to run variety on my home PC and mark images as favourites. Then on my work PC I simply copy over my favourites and only rotate through those (I don't download any new wallpapers at work) That way I've personally vetted each and every wallpaper showing up on my work PC. Its an imperfect solution but it works for me.

I would still like the ability to mark wallpapers as safe vs unsafe OR perhaps use something like https://github.com/yahoo/open_nsfw.

It seems like you should be able to build a database of "safe" vs unsafe imagery relatively easily. Meaning you start with some core "safe" imagery that's been pre-vetted, then you let users score images, once a particular image achieves a certain threshold of approvals its deemed "safe". Over time it should be possible to "build" a library of safe images.

That said I'm probably oversimplifying things greatly :) If it was so simple I'm sure Peter would have implemented it already.

I believe there were also concerns about copyright infringement and downloading imagery from un-vetted sources so there's that angle as well.

Indeed, the original safe mode implementation relied on an image metadata site (vrty.org) that was shut down over a year ago[1][2]. Effectively safe mode was useless since, until the options relying on vrty.org were removed in 0.6.8

[1]: http://peterlevi.com/variety/2017/03/changes/
[2]: https://bugs.launchpad.net/variety/+bug/1635912

Ah, I remember there being announcements about vrty.org. I should have made the connection. I re-read the March 2017 post just now.

When I first posted this, I imagined a simpler person-detector that just filtered *all* images with people in them, rather than making a NSFW judgement. But maybe the yahoo model that Brad mentions would work, though it would require Caffe, which I imagine is a difficult-to-guarantee requirement. Since I think that the cpu-only version of tensorflow would be pretty easy to require (installs in a Python 3 environment with pip only), I tried running it through caffe-tensorflow[1], but that failed. However, if it's just a person filter that we want, maybe the yahoo model wouldn't be needed. And, since speed isn't an issue, maybe it could be deployed with Numpy only.

With a few thousand labeled images, it might be possible to use, say, a small pre-trained VGG network to do this. Peter, does such data exist? Grepping through the source, I see only PIL and cairo as, I think, non-stdlib imports. CouldnNumpy or tensorflow (not tensorflow-gpu) be added?

[1]: https://github.com/ethereon/caffe-tensorflow