curlicuecal:

curlicuecal:

malepresentingnips:

I would say about 50% owl

[colors tweaked post hoc to satisfy the censorbots]

this is the best science I have ever done

@lewisandquark has an excellent post about why this kind of simple tweak can throw off a neural net—

Do Neural Nets Dream of Electric Sheep”

For example, image recognition has a tendency to misindentify rocky fields as goats (because goats are often found there), images with rulers are more likely to contain skin cancer, annnnnd apparently anything significantly skin toned as porn.

( #send dunes)

Not only that, but nets that were pretty good at recognizing goats regularly misIDed them when they turned up in unusual places– goats in trees = birds, goats in cars = dogs

Neural nets inherently reproduce the biases of the systems they’re trained on.

And yeah, playing with the censorbots reveals all sorts of pattern breaks can throw them off— large borders, weird zooms, image filters, hue, and saturation changes, clothing, props.

The presence of a small owl in a hat.

Tl;dr:

friends, may I suggest the solution to the nsfw-ban is porn in unusual places?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s