mirror of
https://github.com/Anjok07/ultimatevocalremovergui.git
synced 2025-02-12 09:02:57 +01:00
Add files via upload
This commit is contained in:
Anjok07
GitHub
parent
85985107bf
commit
d2236f74a8
122
lib/layers_537238KB.py
Normal file
122
lib/layers_537238KB.py
Normal file
@ -0,0 +1,122 @@
|
||||
import torch
|
||||
from torch import nn
|
||||
import torch.nn.functional as F
|
||||
|
||||
from lib import spec_utils
|
||||
|
||||
|
||||
class Conv2DBNActiv(nn.Module):
|
||||
|
||||
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
|
||||
super(Conv2DBNActiv, self).__init__()
|
||||
self.conv = nn.Sequential(
|
||||
nn.Conv2d(
|
||||
nin, nout,
|
||||
kernel_size=ksize,
|
||||
stride=stride,
|
||||
padding=pad,
|
||||
dilation=dilation,
|
||||
bias=False),
|
||||
nn.BatchNorm2d(nout),
|
||||
activ()
|
||||
)
|
||||
|
||||
def __call__(self, x):
|
||||
return self.conv(x)
|
||||
|
||||
|
||||
class SeperableConv2DBNActiv(nn.Module):
|
||||
|
||||
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
|
||||
super(SeperableConv2DBNActiv, self).__init__()
|
||||
self.conv = nn.Sequential(
|
||||
nn.Conv2d(
|
||||
nin, nin,
|
||||
kernel_size=ksize,
|
||||
stride=stride,
|
||||
padding=pad,
|
||||
dilation=dilation,
|
||||
groups=nin,
|
||||
bias=False),
|
||||
nn.Conv2d(
|
||||
nin, nout,
|
||||
kernel_size=1,
|
||||
bias=False),
|
||||
nn.BatchNorm2d(nout),
|
||||
activ()
|
||||
)
|
||||
|
||||
def __call__(self, x):
|
||||
return self.conv(x)
|
||||
|
||||
|
||||
class Encoder(nn.Module):
|
||||
|
||||
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
|
||||
super(Encoder, self).__init__()
|
||||
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
|
||||
self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
|
||||
|
||||
def __call__(self, x):
|
||||
skip = self.conv1(x)
|
||||
h = self.conv2(skip)
|
||||
|
||||
return h, skip
|
||||
|
||||
|
||||
class Decoder(nn.Module):
|
||||
|
||||
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False):
|
||||
super(Decoder, self).__init__()
|
||||
self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
|
||||
self.dropout = nn.Dropout2d(0.1) if dropout else None
|
||||
|
||||
def __call__(self, x, skip=None):
|
||||
x = F.interpolate(x, scale_factor=2, mode='bilinear', align_corners=True)
|
||||
if skip is not None:
|
||||
skip = spec_utils.crop_center(skip, x)
|
||||
x = torch.cat([x, skip], dim=1)
|
||||
h = self.conv(x)
|
||||
|
||||
if self.dropout is not None:
|
||||
h = self.dropout(h)
|
||||
|
||||
return h
|
||||
|
||||
|
||||
class ASPPModule(nn.Module):
|
||||
|
||||
def __init__(self, nin, nout, dilations=(4, 8, 16, 32, 64), activ=nn.ReLU):
|
||||
super(ASPPModule, self).__init__()
|
||||
self.conv1 = nn.Sequential(
|
||||
nn.AdaptiveAvgPool2d((1, None)),
|
||||
Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
|
||||
)
|
||||
self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
|
||||
self.conv3 = SeperableConv2DBNActiv(
|
||||
nin, nin, 3, 1, dilations[0], dilations[0], activ=activ)
|
||||
self.conv4 = SeperableConv2DBNActiv(
|
||||
nin, nin, 3, 1, dilations[1], dilations[1], activ=activ)
|
||||
self.conv5 = SeperableConv2DBNActiv(
|
||||
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ)
|
||||
self.conv6 = SeperableConv2DBNActiv(
|
||||
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ)
|
||||
self.conv7 = SeperableConv2DBNActiv(
|
||||
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ)
|
||||
self.bottleneck = nn.Sequential(
|
||||
Conv2DBNActiv(nin * 7, nout, 1, 1, 0, activ=activ),
|
||||
nn.Dropout2d(0.1)
|
||||
)
|
||||
|
||||
def forward(self, x):
|
||||
_, _, h, w = x.size()
|
||||
feat1 = F.interpolate(self.conv1(x), size=(h, w), mode='bilinear', align_corners=True)
|
||||
feat2 = self.conv2(x)
|
||||
feat3 = self.conv3(x)
|
||||
feat4 = self.conv4(x)
|
||||
feat5 = self.conv5(x)
|
||||
feat6 = self.conv6(x)
|
||||
feat7 = self.conv7(x)
|
||||
out = torch.cat((feat1, feat2, feat3, feat4, feat5, feat6, feat7), dim=1)
|
||||
bottle = self.bottleneck(out)
|
||||
return bottle
|
||||
113
lib/nets_537238KB.py
Normal file
113
lib/nets_537238KB.py
Normal file
@ -0,0 +1,113 @@
|
||||
import torch
|
||||
import numpy as np
|
||||
from torch import nn
|
||||
import torch.nn.functional as F
|
||||
|
||||
from lib import layers_537238KB as layers
|
||||
|
||||
|
||||
class BaseASPPNet(nn.Module):
|
||||
|
||||
def __init__(self, nin, ch, dilations=(4, 8, 16)):
|
||||
super(BaseASPPNet, self).__init__()
|
||||
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
|
||||
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
|
||||
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
|
||||
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
|
||||
|
||||
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
|
||||
|
||||
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
|
||||
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
|
||||
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
|
||||
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
|
||||
|
||||
def __call__(self, x):
|
||||
h, e1 = self.enc1(x)
|
||||
h, e2 = self.enc2(h)
|
||||
h, e3 = self.enc3(h)
|
||||
h, e4 = self.enc4(h)
|
||||
|
||||
h = self.aspp(h)
|
||||
|
||||
h = self.dec4(h, e4)
|
||||
h = self.dec3(h, e3)
|
||||
h = self.dec2(h, e2)
|
||||
h = self.dec1(h, e1)
|
||||
|
||||
return h
|
||||
|
||||
|
||||
class CascadedASPPNet(nn.Module):
|
||||
|
||||
def __init__(self, n_fft):
|
||||
super(CascadedASPPNet, self).__init__()
|
||||
self.stg1_low_band_net = BaseASPPNet(2, 64)
|
||||
self.stg1_high_band_net = BaseASPPNet(2, 64)
|
||||
|
||||
self.stg2_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
|
||||
self.stg2_full_band_net = BaseASPPNet(32, 64)
|
||||
|
||||
self.stg3_bridge = layers.Conv2DBNActiv(130, 64, 1, 1, 0)
|
||||
self.stg3_full_band_net = BaseASPPNet(64, 128)
|
||||
|
||||
self.out = nn.Conv2d(128, 2, 1, bias=False)
|
||||
self.aux1_out = nn.Conv2d(64, 2, 1, bias=False)
|
||||
self.aux2_out = nn.Conv2d(64, 2, 1, bias=False)
|
||||
|
||||
self.max_bin = n_fft // 2
|
||||
self.output_bin = n_fft // 2 + 1
|
||||
|
||||
self.offset = 128
|
||||
|
||||
def forward(self, x, aggressiveness=None):
|
||||
mix = x.detach()
|
||||
x = x.clone()
|
||||
|
||||
x = x[:, :, :self.max_bin]
|
||||
|
||||
bandw = x.size()[2] // 2
|
||||
aux1 = torch.cat([
|
||||
self.stg1_low_band_net(x[:, :, :bandw]),
|
||||
self.stg1_high_band_net(x[:, :, bandw:])
|
||||
], dim=2)
|
||||
|
||||
h = torch.cat([x, aux1], dim=1)
|
||||
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
|
||||
|
||||
h = torch.cat([x, aux1, aux2], dim=1)
|
||||
h = self.stg3_full_band_net(self.stg3_bridge(h))
|
||||
|
||||
mask = torch.sigmoid(self.out(h))
|
||||
mask = F.pad(
|
||||
input=mask,
|
||||
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
|
||||
mode='replicate')
|
||||
|
||||
if self.training:
|
||||
aux1 = torch.sigmoid(self.aux1_out(aux1))
|
||||
aux1 = F.pad(
|
||||
input=aux1,
|
||||
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
|
||||
mode='replicate')
|
||||
aux2 = torch.sigmoid(self.aux2_out(aux2))
|
||||
aux2 = F.pad(
|
||||
input=aux2,
|
||||
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
|
||||
mode='replicate')
|
||||
return mask * mix, aux1 * mix, aux2 * mix
|
||||
else:
|
||||
if aggressiveness:
|
||||
mask[:, :, :aggressiveness['split_bin']] = torch.pow(mask[:, :, :aggressiveness['split_bin']], 1 + aggressiveness['value'] / 3)
|
||||
mask[:, :, aggressiveness['split_bin']:] = torch.pow(mask[:, :, aggressiveness['split_bin']:], 1 + aggressiveness['value'])
|
||||
|
||||
return mask * mix
|
||||
|
||||
def predict(self, x_mag, aggressiveness=None):
|
||||
h = self.forward(x_mag, aggressiveness)
|
||||
|
||||
if self.offset > 0:
|
||||
h = h[:, :, :, self.offset:-self.offset]
|
||||
assert h.size()[3] > 0
|
||||
|
||||
return h
|
||||
Loading…
x
Reference in New Issue
Block a user