reiinakano

but they can all invert a binary tree on a whiteboard, so all good

No, it doesn't. Shared weight means all weights are the same value, as I originally mentioned.

Neither. The architecture performs well independently of weights BUT all weights must be the same e.g. Same network works well when all weights are 5.0 or when all weights are -3.0

> I'm much less confident about everything

This doesn't seem like a bad thing.

Maybe people just realize as they grow older how complex things really are, and eventually learn not to fall for the dunning Kruger effect on everything

I did a search for Flamingo and there were no flamingos...

Cause you can do things like this https://magenta.tensorflow.org/blog/2018/12/20/style-transfe...

Yes, a lot of other apps and websites have been doing this for years. But you have to upload your images to their server. The point of this is that it does all calculations locally in the browser, at the expense of some quality.

That's a good observation. Theoretically, 0 strength should give you the content image. What I think is happening is that the algorithm is not trained to generate style vectors for photorealistic images, and so the mapping it learns from pixels to style vector doesn't work well. Maybe a term should be added to the loss function to generate itself when content and style image are equal.

Hi HN! A year ago, I shared a demo where you could paint a photograph in the style of one of five paintings, via neural networks, right in the browser.

Since then, a few papers have come out attempting to create a single model that could be used for all styles. I tried porting one of those models to the browser today.

A brief summary of how the algorithm works:

- For any particular style, a neural network encodes it into a 100-dimensional vector that represents the network's "understanding" of the style.

- This vector is fed, along with the content image, to another neural network that does the style transformation.

This is also how combining two styles work. The mean of the style vectors of Style A and Style B is calculated and used as the style vector input to the transformation network.

In any case, I acknowledge the results are not perfect and will not look good for all combinations of style and content (particularly faces, ugh), but I think it's a good reason to get excited about what will eventually become possible in the future using the browser alone.

"Theranos, diagnose diseases with AI and Blockchain support"

> I don't think the author has any insight at all on what a general AI will be.

To be fair, nobody does. But you're right, the author shouldn't be making these statements with such certainty.

Sure, they do their best to give an explanation after discovering it works, but nobody ever publishes the hundred other models they tried that don't work out.

Looks about right. The default setting is as basic as you can get. No transpose convolutions, batch norm, and other tricks for better GANs.

Unlike regular classification, it's hard to quantify the performance of a GAN by just looking at the loss curves. You really have to make a subjective evaluation by looking at the generated images.

By playing around with it, the results I got from fully connected were nowhere near as good as the results I got from convolutional.

Default parameters on MNIST should give good results in 5 minutes give or take - depending on your hardware. Can you show what you got?

Yes, most likely.

At most they probably tweaked things a little bit for some minor improvements, but the underlying idea and algorithm is very likely the same.

Can't confirm this, though. :)

Doesn't support mobile :) Please use desktop (Chrome suggested)

To everyone facing this issue, please leave a comment here with your browser and system specs:

https://github.com/reiinakano/fast-style-transfer-deeplearnj...

To everyone facing this issue, please leave a comment here with your browser and system specs:

https://github.com/reiinakano/fast-style-transfer-deeplearnj...

Fast Neural Style transfer is a variation of a "regular" Neural Style algorithm.

With regular Neural Style, you can do style transfer between any two arbitrary images. The disadvantage is it takes longer so doing it in the browser might be unfeasible for now. See https://github.com/anishathalye/neural-style

Anyway, even with Fast Neural Style, you can use any arbitrary image as a Style, but you'll have to train it first (4-6 hours on a GPU).

Dunno about you, but when I first saw some of the results, I was amazed. Here are a few more examples:

https://github.com/jcjohnson/fast-neural-style

https://github.com/lengstrom/fast-style-transfer

Admittedly, though, I haven't seen the Photoshop filters you speak of. Could you link to some of them that show these same effects?

Copying from nature with tiny little variations I guess?

How much do you take? Do you think this is enough for any number of possible use cases in the future e.g. multiple diseases, experimental treatments, etc. Is there a real possibility of running out of banked stem cells?

What are the options for someone living in another country?

Stupid question: Is CloudFlare free for CDN purposes? Right now I'm serving everything from Github Pages. I don't expect there to be much traffic after this initial HN spike dies down, so if it isn't free, it won't be worth it for me to pay for it.

Ah, that makes sense. Thanks a lot for the detailed explanation!

Didn't really use any of their code. I reimplemented their model in a new library called deeplearn.JS.

If the issue is indeed with the library, there's little I can do.

I will still try to find what the issue is but can't make promises. Sorry for the letdown!

I was able to stylize it fine. Maybe refresh and try again with a smaller size?

These definitely look wrong. Unfortunately, it's super hard to debug since I can't reproduce. :(

My only advice I guess is to try it on a different machine when you have one available. Sorry, if this is a problem with the library I used, I can't really do anything to fix it. :/

You're lengstrom from Github! I must let you know your models are exactly what I used! I don't have a beefy GPU so I was forced to use what you already had, haha.

For everyone reading, these guys coded up the neural network I used, all I did was port it to a different library. :)

Eagerly waiting for the Tensorfire API!