Legacy ImageMagick Discussions Archive

I came across this website: www.remove.bg.
It removes backgrounds. OMG. It's like magic.
I threw quite a few images and it gave a SUPER clean cut image with background removed with no dodgy cutting out.

The images that I gave had a beige background on 90% of the image (top part).
The bottom part had floorboard skirting and a lamenate floor which had random patterns.

This is real MAGIC!
Except... magic doesn't exist.
If they can do it on a website, then I know I can get the SAME result using ImageMagick.
+ I'm guessing some clever people have already written many scripts to do the EXACT same with little effort?

Any pointers would be great.

Thanks.

Actually this is a difficult task. It requires some special code that may include artificial intelligence / deep learning. I have never seen anything like that before that does not require some user interaction. So it must be quite complex. I have spent quite some time experimenting with background removal and am quite impressed. But have no idea how they do that.

oh darn it! grrr!

i was hoping there was some other code to match it out there!

i've just spent 10-15 minutes trying other websites.
this websites beats all other hands down. it's pretty awesome.

but i cant still believe there is a 'magic sauce' that no one else has!

It is not that no one else knows how. It is just complicated to implement and requires GPU accelerated processing and deep learning tools. Just do a Google Search on "background removal with deep learning". See for example:

https://towardsdatascience.com/backgrou ... f2104b3157
https://news.developer.nvidia.com/this- ... m-a-photo/
https://www.theverge.com/2018/12/19/181 ... -photoshop
https://github.com/eti-p-doray/unet-gan-matting
https://github.com/SaoYan/bgsCNN
https://www.digitaltrends.com/photograp ... emoval-ai/
https://01.org/zh/node/29971
https://www.ntt-review.jp/archive/nttte ... 12fa2.html

I've been reading those links in the past 10-15 minutes
hmmm... seems as though you were correct.
one of the links is about an article on that website.
the first link is really a good read.
i'll try out the python code links you gave.
THANKS!