Frequency Control does not work

[blackzebra7]

when I ran the command python run2d.py as instructed, I obtained results that were identical to those of DDS. Upon closer inspection of the code, I noticed that the variables keep_low and keep_high do not seem to actually control the editing process as expected.

[delaprada]

when I ran the command python run2d.py as instructed, I obtained results that were identical to those of DDS. Upon closer inspection of the code, I noticed that the variables keep_low and keep_high do not seem to actually control the editing process as expected.

I think so. I have the same problem with both buddha and coffee samples.

[delaprada]

when I ran the command python run2d.py as instructed, I obtained results that were identical to those of DDS. Upon closer inspection of the code, I noticed that the variables keep_low and keep_high do not seem to actually control the editing process as expected.

I think so. I have the same problem with both buddha and coffee samples.

I think the code doesn't include the variables keep_high and keep_low. I fixed it by simply comment:
keep low frequency, comment this code:

FDS/run2d.py

Line 369 in e03c01e

Yl = Yl.requires_grad_(True)

or
keep high frequency, comment this code:

FDS/run2d.py

Line 370 in e03c01e

Yh = [Yh_part.requires_grad_(True) for Yh_part in Yh]

This produces the correct result:

[robot0321]

Thanks to @delaprada,

The result aligned with the paper is from

if not keep_low:
    Yl = Yl.requires_grad_(True) 
if not keep_high:
    Yh = [Yh_part.requires_grad_(True) for Yh_part in Yh] 

, because .requires_grad_(True) activates the target frequency range tensor for change.

It might be confusing whether the pattern on top of the coffee belongs to the high-frequency range, but I think this is simply because the authors arbitrarily defined what counts as “low-frequency” and “high-frequency.” You can modify the code to set a more appropriate frequency cutoff threshold if needed.