GT for cluster and linear probe

[Eric-L-Manibardo]

Hello, I need a table that matches pixel RGB color to label (e.g. [100,0,0] to the sky).

CLUSTERING
From the paper, I know that Hungarian matching is used for aligning the unlabeled clusters with the GT (ground truth). However, I do not understand how to reproduce this step on my own (I have as materials a set of clustered images and the Cityscapes test holdout as GT).

LINEAR
Linear probe uses GT for training a simple NN. Therefore, the matching between real and predicted labels is straightforward. However, I can not find such relationship, neither how to reproduce the process myself.

Thanks in advance

[Eric-L-Manibardo]

Everything is explained on a solved issue at #18
It was a mistake of my code at the time of storing the images where the RGB values were being altered, so I was not able to link RGB to class labels.