Investigate/perform a semantic segmentation of our data using linkNet

[kkmcgg]

LinkNet is a faster variant of a U-net design to perform semantic segmentation using neural networks. Sematic meaning each pixel of the output image is assigned a classification code. Perhaps the best tutorial I've come across to accomplish this task is done by Sreenivas Bhattiprolu on his youtube channel DigitalSreeni. His tutorials are truly the best for quickly implementing neural net functionality!

Specifically - his video comparing U-net and LinkNet image segmentations is very brief and direct. It shows very clearly how to initiate and use both neural net types. He actually uses pre-trained networks accessed an open repository segmentation models.. Using pretrained models greatly reduces the computation time required to get good results. He performs the segmentation with both techniques and minimal code on only a few hundred training data.

Check out this video and the included code to get a good tutorial on training and performing linkNet image segmentation:

Your Task:

• review this video and report back the essential steps which relate to our data
• (optional) attempt to construct a new .ipynb notebook in google colab which demonstrates re-training a linkNet from segmentation_models using 500+ map image tiles collected using our getTile() function.

[kkmcgg]

An example of this has been posted in our experiment section which can be viewed live on google colab.