machine learning for human extraction
Hello guys.I am starting image-processing to extract human bodies from static background properly, I thought big: a general background remover that will automatically identify the foreground and background in every type of image.Now I think it is imposibble with OPENCV.Maybe ,is machine learning useful to extract human body foreground? What do you think about this issue?
what you probably want is semantic segmentation, not background/foreground
then, there's HOGdetectors, Cascades and dpm models, pretrained to detect humans. (they give you the position / boundingrect, not the outline, though)
thank you that's what i am looking for
maybe you could start here , and just use another cascade
Thank you.Is haar cascades and semantic segmentation same thing?I do some research on google and find Neural Network stuff.I saw article about semantic segmentation that can be remove background: https://towardsdatascience.com/backgr...
no not the same. cascades are trained to find specific things (faces, stop signs, eyes) while semantic segmentation divides the image into objects.
btw, the fcn network mentioned in your (quite nice !) article above can be used from opencv, probably even from js (though there is no tutorial or demo code for it)
There is a lot of Semantic Segmentation technique.FCN,SegNet... How can ı learn "what is the difference between these techniques?"
hahahahahaha good one.
read a lot of papers, some courses, lot of experiments ? i know, it all looks overwhelming, and there is no easy answer on "how to catch up".
to quote from the article above:
and that is for someone deep into the subject for years ...
Thanks.It needs to be intermediate understanding of neural networks and deep learning.Is there a tutorial or book for learn semantic segmentation?I want to learn this things hierarchical structure.
another good question.
idk anything specific to segmentation, but my favourite goto for deep learning / nns is here
Thank you a lot.Can i implement these trained models on my program without learning "How to train these networks?"?Is there any avaible trained model for human extraction?I want to learn "Basics of the Semantic Segmentation" and then implement these models on my program.