The new forums will be named Coin Return (based on the most recent vote)! You can check on the status and timeline of the transition to the new forums here.
The Guiding Principles and New Rules document is now in effect.
Minimalist [chat]
Shivahn
Unaware of her barrel shifter privilegeWestern coastal temptressRegistered User, Moderator mod
Unaware of her barrel shifter privilegeWestern coastal temptressRegistered User, Moderator mod
This is a chat. It has everything a chat needs: a title containing the word chat, and a body. I guess it could use a bunch of cat photos to spice things up:
Maybe an extra cute one?
This concludes the body of minimalist chat.
Maybe an extra cute one?
This concludes the body of minimalist chat.
+36
This discussion has been closed.
Posts
It could be many things but a [Chat] is one of the things it must be based on established criteria
Ever wonder what sonar actually sounds like in the water? Here you go!
Pick your alignment I'm sepia woodland
And maybe Mike Mignola's Hellboy art.
it's fuckin' wild. you also can feel it in your body. I remember hearing it a few times while scuba diving.
also whales are fuckin' loud
sperm whales can apparently blow out your ear drum and I can attest to how bone-shaking it is
These are the eyes I most associate with her.
Also, in the interest of being fair to her, this is probably the image that best shows her standard stance, aggression, and level of angelicness:
(I have a ton of cute kitten photos too, but they are for some other time, I am tired of cropping and scaling images)
Look sometimes you need to be able to shout "~How you doin?~ over 1000 miles ok
Basically, what happens if you feed it Ralph Steadman?
Oh, you missed my alignment chart version:
I can control the relative level and scale of distortion, actually, though I haven't actually played with that and thus have little to no skill in that manner of manipulation.
iirc from when I was playing around with style transfer, I think it has to do something with what layer of the CNN you "cross the streams" at?
I'm not exactly sure how Shiv's style transfer works, though.
and the gengars who are guiding me" -- W.S. Merwin
Does she have a twitter account?
I'm using the OG style transfer, the extremely slow image-optimization one by Gatys et al., 2015. So yeah, the level can be controlled based on which layer's gram matrices I use (as well as the content layer). The only adjustment I've made to the algorithm as written in the paper (which is different from what's written in other papers, none of which agree....) is that for speed reasons I've been initializing with the content image instead of random initialization. Also, I'm going to upscale these images later using the technique he suggested in the follow up paper, though it's kind of an obvious idea and I actually realized I could do it before reading the paper.
Which one? The fluffy one has a Himalayan style smooshed face that makes her breathing sound funny.
Lia's is plainer than it looks, she's just real bad at getting flattering images taken. The most interesting thing is something I unfortunately don't have images of off hand, though you can see it in a couple of the photos: her lower mouth is split down the middle wrt color, so she has a black lip on one side and a pink one on the other, and the roof of her mouth is also speckled in that manner, which is the result of a quirk of the intersection of mammalian sex determination and feline coat color genetics.
She probably should, just because there are so many photos of her being a huge dork..
Ah, cool, yeah that's the one I messed with
Re: the upscale thing, I think in general that's where the really cool shit can be done art-wise, by running images through multiple different NNs in order to do multiple operations on an image. Next time I want to try a lot of combinations of passing images through multiple times or through multiple different kinds of nets.
Warframe/Steam: NFyt
Iirc You can do this and it will turn out different. Takes more time to run though ofc.
Warframe/Steam: NFyt
and the gengars who are guiding me" -- W.S. Merwin
I want to go bake in a hot sauna
When the planets align the Great Unclean One will be summoned
I am gonna futz about with other types of networks after doing some more work with this (this is all VGG19).
Probably. It's hard to explain what's going on exactly, but basically:
The neural network is built such that it has many layers, each of which takes previous layers as inputs and looks for features. So the first layer might look for lines (vertical, horizontal, other) or color or whatever. Next layer might look for combinations of these features (vertical AND horizontal lines? This is a hash mark; or, lines angled up-left , vertical, and then down-right? Maybe that's a circle). They get more and more complicated until it's doing things like recognizing eyes or cats (or orange cats, or animal ears, or cars - you get the idea).
The algorithm works by taking two images - a content image and a style image. It tries to match the content image exactly, at whatever layer you say (so, if you try to match to an early layer, it'll VERY faithfully re-create the original image, because it's trying to keep the image as close as possible wrt lines, colors, dots, etc, but if you choose a later layer it'll be matching more vaguely - trying to make sure it remains a cat, with gray stripes, on a tan cat tower, etc).
The style image it tries to match correlations - so, taking the Wave for example, it'll pick up on the fact that there's a lot of white and a lot of blue, but not both in the same place, and when they meet it is usually at curves, and it'll try to make curvy white/blue separations.
Anyway. You could absolutely try to match it to multiple images - you'd just say "hey, instead of matching the correlations to the Wave, map them to the Wave and the Dream of the Fisherman's Wife." You'd probably halve the strength of each image, since you don't want to overwhelm it. If you matched it to ALL of Hokusai's work, it'd try to most faithfully represent the average "style" seen in those works.
Hmm that is not a great explanation, but it's really hard to find a good lay way of explaining it when it boils down to "this particular form of a correlationary matrix seems to work for this task."