What to use for Semantic Segmentation

Hello, just to briefly introduce myself, I am just starting to work with deep learning and I don't really know much about it. I have a task that I want to complete and I would very much appreciate to be given some sort of a direction about what should I start to look at and learn about.

My task is to perform semantic segmentation on images like the following:

https://drive.google.com/open?id=1G9LTJB3BP1aWF8xutix6YxjSITtSRH-a

I have about 1500 similar images as a training set, which are not pixel label classified. The task is the be able to get information about where on a new image are there structures like the ones marked with green.

Should I look at some premade neural network architectures for semantic segmentation, or would I need something custom. Also, do I have to create pixel labels for the images by hand to train a network, or is there a more automatic approach?

Answer 1

[Sebastian Castro]

You can start with premade neural network architectures. The following example does this with the VGG-16 architecture:

https://www.mathworks.com/help/vision/examples/semantic-segmentation-using-deep-learning.html

As far as labeling the pixel data, check this out:

https://www.mathworks.com/help/vision/ug/label-pixels-for-semantic-segmentation.html

- Sebastian