De-standarization of output data

[luiservela]

Dear all,
I have been looking at the inference.py script and I see how lines 190 and 191 (

FourCastNet/inference/inference.py

Lines 190 to 191 in 93360c1

	valid_data = (valid_data - means)/stds
	valid_data = torch.as_tensor(valid_data).to(device, dtype=torch.float)

) in the autoregresive_interpolation function help out standardize the input data.

However, I see nowhere the inverse process taking place for output data (seq_real and seq_pred).

Could this be achieved by simply running:
seq_real = seq_real * stds + means
seq_pred = seq_pred * stds + means

or am I missing something critical here?

Best,

LuisV

[Mo-Elfeky]

Did you find the right solution ?
i'm walking on your path, bcz there's a lot of errors exist in this code

[wangjiayiiiii]

Excuse me,
some data from the NVlabs/FourCastNet project now seems to be closed-source.
Do you still have the stats_v0 data in normalization?