Mixup: changes in labels when combining images in different classes

Hi all,
I want to use mixup as the first step of my augmentation data, and I’m scared about the impact of this in my labels.
Can I assume that : t * img1 + (1-t) * img2 => t * y1 + (1-t) * y2 regardless of the model I’m using ? In what context could this be true (without degrading the performance of my model)