Why are FP and FN different for each Tracker method in the paper?

[YoungjaeDev]

As you can see below, they used the same FrRCNN (VGG16) model in the paper and trained with the same dataset, but why do they get different detection FP and FN each other?