This is a quick proof of concept of an idea. This won't work for every use-case but it worked surprisingly well for a use-case I had in mind. Happy to answer any questions.
So, is the ML part just the segmentation, correct?
Yes. That is correct.
Bokehlicious Selfies
11–20 of 53 posts
Re: Bokehlicious Selfies
#12This is really cool, but it isn't bokeh, it's just blur.
Yeah. I don’t see anything here that couldn’t be done in Photoshop with a bit of masking work.
Here’s an example [1] of how bokeh really looks, using a high quality fast lens on a decent camera. We’ve got a long way to go.
[1] https://upload.wikimedia.org/wikipedia/commons/8/8a/Josefina...
Re: Bokehlicious Selfies
#13I just discovered that Zoom teleconferencing software has the ability to detect the background and replace it with an arbitrary video in real time. I have no idea how they do it, but it's as impressive as heck.
Re: Bokehlicious Selfies
#14OT: I can't understand why this website _want_ to looks like a mobile web __and__ hides scrolling bar. I'm on desktop and it irritates me a bit.
Re: Bokehlicious Selfies
#15One step further would be to incorporate the blurring, masking, and combining operations into the network itself, so you can leverage the GPU to do the whole computation in one fell swoop.
Re: Bokehlicious Selfies
#16OT: I can't understand why this website _want_ to looks like a mobile web __and__ hides scrolling bar. I'm on desktop and it irritates me a bit.
Might be that it's a mobile first theme that's more like mobile only – styling for the phone and not making any changes for desktop.
Re: Bokehlicious Selfies
#17This is really cool, but it isn't bokeh, it's just blur.
Yeah. I don’t see anything here that couldn’t be done in Photoshop with a bit of masking work. Here’s an example [1] of how bokeh really looks, using a high quality fast lens on a decent camera. We’ve got a long way to go. [1] https://upload.wikimedia.org/wikipedia/commons/8/8a/Josefina...
The author also uses a neural network to automatically detect the foreground and apply the filter only to the background so this can be done automatically.
Re: Bokehlicious Selfies
#18Wow this looks ugly :D. But still nice work. It's just that the girl is totally over-processed and the "bokeh", well I hate to break it to you, but this isn't bokeh, its called "blur".
This is bokeh: https://cdn.mos.cms.futurecdn.net/JgrDuxQPCAvgd5VzqiKN5a-650...
And bokeh is 3 dimensional and surrounds the focus area, increasing with distance: https://www.flickr.com/photos/scottcartwrightphotography/145...
I suggest to brush up a bit on photo skills, although I agree that the instragram and iPhone culture can shift your perception of "bokehlicious" quite a bit.
I don't want to make you feel bad, so I also tell you that Apple's "bokeh" in portrait mode looks disgusting, doesn't make any sense and breaks the image in a lot of trivial cases... and they spent a lot more money & effort on it than you :).
Re: Bokehlicious Selfies
#19Wow this looks ugly :D. But still nice work. It's just that the girl is totally over-processed and the "bokeh", well I hate to break it to you, but this isn't bokeh, its called "blur". This is bokeh: https://cdn.mos.cms.futurecdn.net/JgrDuxQPCAvgd5VzqiKN5a-650... And bokeh is 3 dimensional and surrounds the focus area, increasing with distance: https://www.flickr.com/photos/scottcartwrightphotography/145... I suggest…
I agree. My goal here was to get something that look like bokeh if you squinted at it.
I just wanted to see how far I could take this idea.
Re: Bokehlicious Selfies
#20I just discovered that Zoom teleconferencing software has the ability to detect the background and replace it with an arbitrary video in real time. I have no idea how they do it, but it's as impressive as heck.
Having just used it, it feels like a few factors:
- a background scene is usually static. if a pixel matches its historical average (i e. not changing) then it's probably background.
- if a pixel color matches its neighbors, it's probably the same as them. i noticed my black shirt tripped it up on occasion, but was all-or-nothing.
- people are blobs, not diffuse, so try to segment large regions.