Create a ground truth dataset of manually labelled segmentation of real RBCs to measure the performance of the segmentation output of the trained neural network. Using some paint software, manually color in each pixel according to it being part of the cell (label=1) or the background (label=0). Before starting, a decision should be made about the definition of cell borders while in the channels, as they often blend into each other, thus ensuring consistency. The expected number of observations in this dataset is 100.
The goal of this issue is twofold:
Prove that the neural network segmentation output works as expected
Prove that the neural network segmentation approach is better than the conventional approaches (#3 )
Create a ground truth dataset of manually labelled segmentation of real RBCs to measure the performance of the segmentation output of the trained neural network. Using some paint software, manually color in each pixel according to it being part of the cell (
label=1
) or the background (label=0
). Before starting, a decision should be made about the definition of cell borders while in the channels, as they often blend into each other, thus ensuring consistency. The expected number of observations in this dataset is 100.The goal of this issue is twofold: