I have a classification map with three classes. I have also a ground truth dataset provided by ROI tools. The numbers of ground truths (pixels) for classes are 870, 1007, and 695. However, when making the confusion matrix to evaluate the classification, it shows different numbers as the ground truth per class i.e. 598, 887, 695. Just for the third class, it is the same.
I do not understand the difference. Why is the total number of pixels of ground truth in the confusion matrix per class different? This would spoil the classification accuracy and all the numbers provided in the report of the confusion matrix.
I am looking forward to hearing your explanation.