mirror of
https://github.com/firestar5683/StarPilot.git
synced 2026-10-08 15:24:14 +08:00
firestar5683
14 lines
858 B
Python
14 lines
858 B
Python
"""Portable exact Haar DCW algebra for diagnostic sampler parity."""
|
|
import numpy as np
|
|
|
|
def dcw(x_next,denoised,t,settings):
|
|
if not settings.get('dcw_enabled',False):return x_next
|
|
assert settings.get('dcw_wavelet','haar')=='haar'
|
|
mode=settings.get('dcw_mode','double');n=x_next.shape[1];x=x_next.astype(np.float32);y=denoised.astype(np.float32)
|
|
if n%2:x=np.pad(x,((0,0),(0,1),(0,0)));y=np.pad(y,((0,0),(0,1),(0,0)))
|
|
s=np.float32(2**-.5);xl=(x[:,::2]+x[:,1::2])*s;xh=(x[:,::2]-x[:,1::2])*s;yl=(y[:,::2]+y[:,1::2])*s;yh=(y[:,::2]-y[:,1::2])*s;scale=settings.get('dcw_scaler',.05)
|
|
if mode in ('low','double'):xl+=t*scale*(xl-yl)
|
|
if mode in ('high','double'):xh+=(1-t)*(settings.get('dcw_high_scaler',.02) if mode=='double' else scale)*(xh-yh)
|
|
assert mode in ('low','high','double')
|
|
return np.stack(((xl+xh)*s,(xl-xh)*s),axis=2).reshape(x.shape)[:,:n]
|