ENH [low priority] option to drop low-pass filters #445
Comments
|
This will cause a huge increase in memory footprint, and might break compatibility if whatever downstream application assumes that all coefficients have the same resolution. Unless you want to oversample? |
|
Oversampling might be OK. Though I'd probably also want to support the
appropriate subsampling.
I am going to write an example that does this and show that I have to
basically copy out the whole loop that does the scattering and do a minor
modification. Given that example, if more people want this as a primary
functionality, we can then consider implementing it later.
Since this codebase is primarily useful for research purposes, in general I
think it would be nice to be able to tinker with the inner states of the
system easily. I get the memory footprint issue, but I think that only
becomes a big deal at layer 2
…On Tue, Oct 29, 2019 at 5:53 PM Vincent Lostanlen ***@***.***> wrote:
This will cause a huge increase in memory footprint, and might break
compatibility if whatever downstream application assumes that all
coefficients have the same resolution. Unless you want to oversample?
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#445?email_source=notifications&email_token=AAJ7AC2GJ6UIYFWBII6MFC3QRDLIRA5CNFSM4JE2SZ42YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOECSR66Y#issuecomment-547692411>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAJ7AC7XJQS3EF2DU345DFDQRDLIRANCNFSM4JE2SZ4Q>
.
|
Yes, this is what prompted me to make |
|
In terms of sampling, it could simply be a matter of disallowing |
|
Does this dictionary stuff we discussed allow this? |
|
@MuawizChaudhary yes, this is the long-term goal of the dictionary output |
I am finding myself in a situation where I need access to the non-lowpass-filtered wavelet moduli.
I realize there is a potential space issue, but it might be not only me wanting access to these.
Going to just convolve the filters myself for the moment, but I might make a PR in the future
The text was updated successfully, but these errors were encountered: