How a robust background removal is implemented?

Question

I found that a deep-learning-based method (e.g., 1) is much more robust than a non-deep-learning-based method (e.g., 2, using OpenCV).

In the OpenCV example, Canny is used to detect the edges. But this step can be very sensitive to the image. The contour detection may end up with wrong contours. It is also difficult to determine which contours should be kept.

How a robust deep-learning method is implemented? Is any good example code? Thanks.

prosti · Answer 1 · 2019-09-15T15:47:42.113

2

For that to work you need to use Unet. You can search for that on github. Unet transofrm is: I->I. Space of the image will become image (of same or similar size).

You need to have say 10.000 images with bg removed. People, (long hair people), cats, cars, shoes, T-shirts, etc.

So you set different backgrounds on all these images as source and prediction should be images with removed background.

You can also do a segmentation model and when you find the foreground you can remove the bg.

edited Sep 15 '19 at 15:47

answered Sep 15 '19 at 14:57

prosti

42,291
14
186
151

Which Unet implementation do you recommend? Is there a trained model ready to use? Thanks. – user1424739 Sep 15 '19 at 15:21
Currently, PyTorch doesn't have Unet pretrained. I would stay neutral. – prosti Sep 15 '19 at 16:30
What about keras or tensorflow? – user1424739 Sep 15 '19 at 16:31
This is a U-Net implementation that could be of benefit: https://github.com/abderhasan/unet – Simplicity Sep 24 '19 at 14:55

How a robust background removal is implemented?

1 Answers1