Why use the original mean and var of each patch when visualizing the reconstruction video?

[PeisenZhao]

When I set a high mask ratio, some unpredictable content will be roughly predicted.

[congee524]

Because the target of reconstruction task is the normalized pixel instead of pixel