DeepArt.io introduces style transfer for videos | Hacker News Reader

DeepArt.io introduces style transfer for videos

deepart.io

65 pointseDameXxX19 comments

It bases on:

- video: https://www.youtube.com/watch?v=Khuj4ASldmU

- paper: Manuel Ruder, Alexey Dosovitskiy, Thomas Brox, "Artistic style transfer for videos" https://arxiv.org/abs/1604.08610

- code: https://github.com/manuelruder/artistic-videos

Thanks! I'm usually way more interested in the technique than in some service offerin it. At least for interesting and novel ideas

At 1k EUR for a 5 minute video, I can't imagine they're getting much in the way of sales.

If it looks like what the client is looking for (a very big if), those are very cheap frames. If you hire someone to manually mangle 5 min of video it will usually end up costing way more than that.

Yeah, that worried me. The pessimist in me suspects that maybe they don't want to make a lot of money selling it, maybe they just want this hypothetical price point to exist when they go out suing people doing similar stuff (the deepart.io people have patented some of the early research in style transfer).

At this price point, it's time to start looking at alternative approaches. The original iterative optimization approach is not the only, nor is it remotely the fastest. (There's an upscaling CNN version which runs at like 20FPS but I can't seem to refind the citation at the moment.)

Especially when their costs are so low?

I've run neural-style on my GPU at home and each frame at high res took ~ 15 minutes to converge. So 5 * 60 * 24 * 15 = 1800 Hours GPU Time * 65 cents an hour on EC2 = $1170. So costs aren't that low.

Try this, should be near realtime

https://arxiv.org/abs/1603.08155

IIRC it takes around a week to pre-compute the style model.

Today I learned that I could be making a lot more money on the side doing this for half the cost with a better method.

I wonder if there are ways to reduce the GPU RAM requirements on this. Even the commercial offering doesn't go beyond 720p - I guess that fills up 12GB on a Titan X?

On stills style transfer you can process the image in multiple tiles, but I don't expect that would work with video with the temporal consistency

   University of Tübingen has a pending patent application for the Neural Art technology.

I wonder what that is (can't find on Google)?

https://news.ycombinator.com/item?id=11873806

It's not dupe! (The other post is on transferring style not color in images; this post is a about transferring style (and color) in videos.)

Ok, we'll undupify it.

I'm pretty certain I saw this exact set of videos posted on another article yesterday, however. Edit: My bad, it was via a youtube link I saw elsewhere, thats also been posted in this thread.

How big companies scale dnn as Google Photos or Siri? Any approach? you need to train big networks for different users?

Watching those videos is how I always imagined what being on LSD would be like.

Reply on news.ycombinator.com