optimize(uvr5): remove redundant files

infer/lib/uvr5_pack/lib_v5/layers.py

@@ -26,40 +26,17 @@ class Conv2DBNActiv(nn.Module):
         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.conv1 = Conv2DBNActiv(nin, nout, ksize, stride, pad, activ=activ)
-        self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, 1, pad, activ=activ)
 
     def __call__(self, x):
-        skip = self.conv1(x)
+        h = self.conv1(x)
-        h = self.conv2(skip)
+        h = self.conv2(h)
 
-        return h, skip
+        return h
 
 
 class Decoder(nn.Module):
@@ -67,15 +44,19 @@ class Decoder(nn.Module):
         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.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        # self.conv2 = Conv2DBNActiv(nout, 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)
+
+        h = self.conv1(x)
+        # h = self.conv2(h)
 
         if self.dropout is not None:
             h = self.dropout(h)
@@ -84,25 +65,24 @@ class Decoder(nn.Module):
 
 
 class ASPPModule(nn.Module):
-    def __init__(self, nin, nout, dilations=(4, 8, 16), activ=nn.ReLU):
+    def __init__(self, nin, nout, dilations=(4, 8, 12), activ=nn.ReLU, dropout=False):
         super(ASPPModule, self).__init__()
         self.conv1 = nn.Sequential(
             nn.AdaptiveAvgPool2d((1, None)),
-            Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
+            Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ),
         )
-        self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
+        self.conv2 = Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ)
-        self.conv3 = SeperableConv2DBNActiv(
+        self.conv3 = Conv2DBNActiv(
-            nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
+            nin, nout, 3, 1, dilations[0], dilations[0], activ=activ
         )
-        self.conv4 = SeperableConv2DBNActiv(
+        self.conv4 = Conv2DBNActiv(
-            nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
+            nin, nout, 3, 1, dilations[1], dilations[1], activ=activ
         )
-        self.conv5 = SeperableConv2DBNActiv(
+        self.conv5 = Conv2DBNActiv(
-            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+            nin, nout, 3, 1, dilations[2], dilations[2], activ=activ
-        )
-        self.bottleneck = nn.Sequential(
-            Conv2DBNActiv(nin * 5, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
         )
+        self.bottleneck = Conv2DBNActiv(nout * 5, nout, 1, 1, 0, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
 
     def forward(self, x):
         _, _, h, w = x.size()
@@ -114,5 +94,32 @@ class ASPPModule(nn.Module):
         feat4 = self.conv4(x)
         feat5 = self.conv5(x)
         out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
-        bottle = self.bottleneck(out)
+        out = self.bottleneck(out)
-        return bottle
+
+        if self.dropout is not None:
+            out = self.dropout(out)
+
+        return out
+
+
+class LSTMModule(nn.Module):
+    def __init__(self, nin_conv, nin_lstm, nout_lstm):
+        super(LSTMModule, self).__init__()
+        self.conv = Conv2DBNActiv(nin_conv, 1, 1, 1, 0)
+        self.lstm = nn.LSTM(
+            input_size=nin_lstm, hidden_size=nout_lstm // 2, bidirectional=True
+        )
+        self.dense = nn.Sequential(
+            nn.Linear(nout_lstm, nin_lstm), nn.BatchNorm1d(nin_lstm), nn.ReLU()
+        )
+
+    def forward(self, x):
+        N, _, nbins, nframes = x.size()
+        h = self.conv(x)[:, 0]  # N, nbins, nframes
+        h = h.permute(2, 0, 1)  # nframes, N, nbins
+        h, _ = self.lstm(h)
+        h = self.dense(h.reshape(-1, h.size()[-1]))  # nframes * N, nbins
+        h = h.reshape(nframes, N, 1, nbins)
+        h = h.permute(1, 2, 3, 0)
+
+        return h

infer/lib/uvr5_pack/lib_v5/layers_123812KB .py

@@ -1,118 +0,0 @@
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from . 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), 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.bottleneck = nn.Sequential(
-            Conv2DBNActiv(nin * 5, 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)
-        out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
-        bottle = self.bottleneck(out)
-        return bottle

infer/lib/uvr5_pack/lib_v5/layers_537227KB.py

@@ -1,126 +0,0 @@
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from . 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

infer/lib/uvr5_pack/lib_v5/layers_new.py

@@ -1,125 +0,0 @@
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from . 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 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, stride, pad, activ=activ)
-        self.conv2 = Conv2DBNActiv(nout, nout, ksize, 1, pad, activ=activ)
-
-    def __call__(self, x):
-        h = self.conv1(x)
-        h = self.conv2(h)
-
-        return h
-
-
-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.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
-        # self.conv2 = Conv2DBNActiv(nout, 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.conv1(x)
-        # h = self.conv2(h)
-
-        if self.dropout is not None:
-            h = self.dropout(h)
-
-        return h
-
-
-class ASPPModule(nn.Module):
-    def __init__(self, nin, nout, dilations=(4, 8, 12), activ=nn.ReLU, dropout=False):
-        super(ASPPModule, self).__init__()
-        self.conv1 = nn.Sequential(
-            nn.AdaptiveAvgPool2d((1, None)),
-            Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ),
-        )
-        self.conv2 = Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ)
-        self.conv3 = Conv2DBNActiv(
-            nin, nout, 3, 1, dilations[0], dilations[0], activ=activ
-        )
-        self.conv4 = Conv2DBNActiv(
-            nin, nout, 3, 1, dilations[1], dilations[1], activ=activ
-        )
-        self.conv5 = Conv2DBNActiv(
-            nin, nout, 3, 1, dilations[2], dilations[2], activ=activ
-        )
-        self.bottleneck = Conv2DBNActiv(nout * 5, nout, 1, 1, 0, activ=activ)
-        self.dropout = nn.Dropout2d(0.1) if dropout else None
-
-    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)
-        out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
-        out = self.bottleneck(out)
-
-        if self.dropout is not None:
-            out = self.dropout(out)
-
-        return out
-
-
-class LSTMModule(nn.Module):
-    def __init__(self, nin_conv, nin_lstm, nout_lstm):
-        super(LSTMModule, self).__init__()
-        self.conv = Conv2DBNActiv(nin_conv, 1, 1, 1, 0)
-        self.lstm = nn.LSTM(
-            input_size=nin_lstm, hidden_size=nout_lstm // 2, bidirectional=True
-        )
-        self.dense = nn.Sequential(
-            nn.Linear(nout_lstm, nin_lstm), nn.BatchNorm1d(nin_lstm), nn.ReLU()
-        )
-
-    def forward(self, x):
-        N, _, nbins, nframes = x.size()
-        h = self.conv(x)[:, 0]  # N, nbins, nframes
-        h = h.permute(2, 0, 1)  # nframes, N, nbins
-        h, _ = self.lstm(h)
-        h = self.dense(h.reshape(-1, h.size()[-1]))  # nframes * N, nbins
-        h = h.reshape(nframes, N, 1, nbins)
-        h = h.permute(1, 2, 3, 0)
-
-        return h

infer/lib/uvr5_pack/lib_v5/nets.py

@@ -1,85 +1,100 @@
-import layers
 import torch
 import torch.nn.functional as F
 from torch import nn
 
-from . import spec_utils
+from . import layers
 
 
-class BaseASPPNet(nn.Module):
+class BaseNet(nn.Module):
-    def __init__(self, nin, ch, dilations=(4, 8, 16)):
+    def __init__(
-        super(BaseASPPNet, self).__init__()
+        self, nin, nout, nin_lstm, nout_lstm, dilations=((4, 2), (8, 4), (12, 6))
-        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+    ):
-        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        super(BaseNet, self).__init__()
-        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc1 = layers.Conv2DBNActiv(nin, nout, 3, 1, 1)
-        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+        self.enc2 = layers.Encoder(nout, nout * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(nout * 2, nout * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(nout * 4, nout * 6, 3, 2, 1)
+        self.enc5 = layers.Encoder(nout * 6, nout * 8, 3, 2, 1)
 
-        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+        self.aspp = layers.ASPPModule(nout * 8, nout * 8, dilations, dropout=True)
 
-        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec4 = layers.Decoder(nout * (6 + 8), nout * 6, 3, 1, 1)
-        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec3 = layers.Decoder(nout * (4 + 6), nout * 4, 3, 1, 1)
-        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec2 = layers.Decoder(nout * (2 + 4), nout * 2, 3, 1, 1)
-        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+        self.lstm_dec2 = layers.LSTMModule(nout * 2, nin_lstm, nout_lstm)
+        self.dec1 = layers.Decoder(nout * (1 + 2) + 1, nout * 1, 3, 1, 1)
 
     def __call__(self, x):
-        h, e1 = self.enc1(x)
+        e1 = self.enc1(x)
-        h, e2 = self.enc2(h)
+        e2 = self.enc2(e1)
-        h, e3 = self.enc3(h)
+        e3 = self.enc3(e2)
-        h, e4 = self.enc4(h)
+        e4 = self.enc4(e3)
+        e5 = self.enc5(e4)
 
-        h = self.aspp(h)
+        h = self.aspp(e5)
 
         h = self.dec4(h, e4)
         h = self.dec3(h, e3)
         h = self.dec2(h, e2)
+        h = torch.cat([h, self.lstm_dec2(h)], dim=1)
         h = self.dec1(h, e1)
 
         return h
 
 
-class CascadedASPPNet(nn.Module):
+class CascadedNet(nn.Module):
-    def __init__(self, n_fft):
+    def __init__(self, n_fft, nout=32, nout_lstm=128):
-        super(CascadedASPPNet, self).__init__()
+        super(CascadedNet, self).__init__()
-        self.stg1_low_band_net = BaseASPPNet(2, 16)
-        self.stg1_high_band_net = BaseASPPNet(2, 16)
-
-        self.stg2_bridge = layers.Conv2DBNActiv(18, 8, 1, 1, 0)
-        self.stg2_full_band_net = BaseASPPNet(8, 16)
-
-        self.stg3_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
-        self.stg3_full_band_net = BaseASPPNet(16, 32)
-
-        self.out = nn.Conv2d(32, 2, 1, bias=False)
-        self.aux1_out = nn.Conv2d(16, 2, 1, bias=False)
-        self.aux2_out = nn.Conv2d(16, 2, 1, bias=False)
 
         self.max_bin = n_fft // 2
         self.output_bin = n_fft // 2 + 1
+        self.nin_lstm = self.max_bin // 2
+        self.offset = 64
 
-        self.offset = 128
+        self.stg1_low_band_net = nn.Sequential(
+            BaseNet(2, nout // 2, self.nin_lstm // 2, nout_lstm),
+            layers.Conv2DBNActiv(nout // 2, nout // 4, 1, 1, 0),
+        )
 
-    def forward(self, x, aggressiveness=None):
+        self.stg1_high_band_net = BaseNet(
-        mix = x.detach()
+            2, nout // 4, self.nin_lstm // 2, nout_lstm // 2
-        x = x.clone()
+        )
 
+        self.stg2_low_band_net = nn.Sequential(
+            BaseNet(nout // 4 + 2, nout, self.nin_lstm // 2, nout_lstm),
+            layers.Conv2DBNActiv(nout, nout // 2, 1, 1, 0),
+        )
+        self.stg2_high_band_net = BaseNet(
+            nout // 4 + 2, nout // 2, self.nin_lstm // 2, nout_lstm // 2
+        )
+
+        self.stg3_full_band_net = BaseNet(
+            3 * nout // 4 + 2, nout, self.nin_lstm, nout_lstm
+        )
+
+        self.out = nn.Conv2d(nout, 2, 1, bias=False)
+        self.aux_out = nn.Conv2d(3 * nout // 4, 2, 1, bias=False)
+
+    def forward(self, x):
         x = x[:, :, : self.max_bin]
 
         bandw = x.size()[2] // 2
-        aux1 = torch.cat(
+        l1_in = x[:, :, :bandw]
-            [
+        h1_in = x[:, :, bandw:]
-                self.stg1_low_band_net(x[:, :, :bandw]),
+        l1 = self.stg1_low_band_net(l1_in)
-                self.stg1_high_band_net(x[:, :, bandw:]),
+        h1 = self.stg1_high_band_net(h1_in)
-            ],
+        aux1 = torch.cat([l1, h1], dim=2)
-            dim=2,
-        )
 
-        h = torch.cat([x, aux1], dim=1)
+        l2_in = torch.cat([l1_in, l1], dim=1)
-        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+        h2_in = torch.cat([h1_in, h1], dim=1)
+        l2 = self.stg2_low_band_net(l2_in)
+        h2 = self.stg2_high_band_net(h2_in)
+        aux2 = torch.cat([l2, h2], dim=2)
 
-        h = torch.cat([x, aux1, aux2], dim=1)
+        f3_in = torch.cat([x, aux1, aux2], dim=1)
-        h = self.stg3_full_band_net(self.stg3_bridge(h))
+        f3 = self.stg3_full_band_net(f3_in)
 
-        mask = torch.sigmoid(self.out(h))
+        mask = torch.sigmoid(self.out(f3))
         mask = F.pad(
             input=mask,
             pad=(0, 0, 0, self.output_bin - mask.size()[2]),
@@ -87,37 +102,32 @@ class CascadedASPPNet(nn.Module):
         )
 
         if self.training:
-            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux = torch.cat([aux1, aux2], dim=1)
-            aux1 = F.pad(
+            aux = torch.sigmoid(self.aux_out(aux))
-                input=aux1,
+            aux = F.pad(
-                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                input=aux,
+                pad=(0, 0, 0, self.output_bin - aux.size()[2]),
                 mode="replicate",
             )
-            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            return mask, aux
-            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:
+            return mask
-                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_mask(self, x):
-
+        mask = self.forward(x)
-    def predict(self, x_mag, aggressiveness=None):
-        h = self.forward(x_mag, aggressiveness)
 
         if self.offset > 0:
-            h = h[:, :, :, self.offset : -self.offset]
+            mask = mask[:, :, :, self.offset : -self.offset]
-            assert h.size()[3] > 0
+            assert mask.size()[3] > 0
 
-        return h
+        return mask
+
+    def predict(self, x, aggressiveness=None):
+        mask = self.forward(x)
+        pred_mag = x * mask
+
+        if self.offset > 0:
+            pred_mag = pred_mag[:, :, :, self.offset : -self.offset]
+            assert pred_mag.size()[3] > 0
+
+        return pred_mag

infer/lib/uvr5_pack/lib_v5/nets_123812KB.py

@@ -1,122 +0,0 @@
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from . import layers_123821KB 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, 32)
-        self.stg1_high_band_net = BaseASPPNet(2, 32)
-
-        self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
-        self.stg2_full_band_net = BaseASPPNet(16, 32)
-
-        self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
-        self.stg3_full_band_net = BaseASPPNet(32, 64)
-
-        self.out = nn.Conv2d(64, 2, 1, bias=False)
-        self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
-        self.aux2_out = nn.Conv2d(32, 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

infer/lib/uvr5_pack/lib_v5/nets_537227KB.py

@@ -1,123 +0,0 @@
-import numpy as np
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from . 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

infer/lib/uvr5_pack/lib_v5/nets_61968KB.py

@@ -1,122 +0,0 @@
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from . import layers_123821KB 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, 32)
-        self.stg1_high_band_net = BaseASPPNet(2, 32)
-
-        self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
-        self.stg2_full_band_net = BaseASPPNet(16, 32)
-
-        self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
-        self.stg3_full_band_net = BaseASPPNet(32, 64)
-
-        self.out = nn.Conv2d(64, 2, 1, bias=False)
-        self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
-        self.aux2_out = nn.Conv2d(32, 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

infer/lib/uvr5_pack/lib_v5/nets_new.py

@@ -1,133 +0,0 @@
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from . import layers_new
-
-
-class BaseNet(nn.Module):
-    def __init__(
-        self, nin, nout, nin_lstm, nout_lstm, dilations=((4, 2), (8, 4), (12, 6))
-    ):
-        super(BaseNet, self).__init__()
-        self.enc1 = layers_new.Conv2DBNActiv(nin, nout, 3, 1, 1)
-        self.enc2 = layers_new.Encoder(nout, nout * 2, 3, 2, 1)
-        self.enc3 = layers_new.Encoder(nout * 2, nout * 4, 3, 2, 1)
-        self.enc4 = layers_new.Encoder(nout * 4, nout * 6, 3, 2, 1)
-        self.enc5 = layers_new.Encoder(nout * 6, nout * 8, 3, 2, 1)
-
-        self.aspp = layers_new.ASPPModule(nout * 8, nout * 8, dilations, dropout=True)
-
-        self.dec4 = layers_new.Decoder(nout * (6 + 8), nout * 6, 3, 1, 1)
-        self.dec3 = layers_new.Decoder(nout * (4 + 6), nout * 4, 3, 1, 1)
-        self.dec2 = layers_new.Decoder(nout * (2 + 4), nout * 2, 3, 1, 1)
-        self.lstm_dec2 = layers_new.LSTMModule(nout * 2, nin_lstm, nout_lstm)
-        self.dec1 = layers_new.Decoder(nout * (1 + 2) + 1, nout * 1, 3, 1, 1)
-
-    def __call__(self, x):
-        e1 = self.enc1(x)
-        e2 = self.enc2(e1)
-        e3 = self.enc3(e2)
-        e4 = self.enc4(e3)
-        e5 = self.enc5(e4)
-
-        h = self.aspp(e5)
-
-        h = self.dec4(h, e4)
-        h = self.dec3(h, e3)
-        h = self.dec2(h, e2)
-        h = torch.cat([h, self.lstm_dec2(h)], dim=1)
-        h = self.dec1(h, e1)
-
-        return h
-
-
-class CascadedNet(nn.Module):
-    def __init__(self, n_fft, nout=32, nout_lstm=128):
-        super(CascadedNet, self).__init__()
-
-        self.max_bin = n_fft // 2
-        self.output_bin = n_fft // 2 + 1
-        self.nin_lstm = self.max_bin // 2
-        self.offset = 64
-
-        self.stg1_low_band_net = nn.Sequential(
-            BaseNet(2, nout // 2, self.nin_lstm // 2, nout_lstm),
-            layers_new.Conv2DBNActiv(nout // 2, nout // 4, 1, 1, 0),
-        )
-
-        self.stg1_high_band_net = BaseNet(
-            2, nout // 4, self.nin_lstm // 2, nout_lstm // 2
-        )
-
-        self.stg2_low_band_net = nn.Sequential(
-            BaseNet(nout // 4 + 2, nout, self.nin_lstm // 2, nout_lstm),
-            layers_new.Conv2DBNActiv(nout, nout // 2, 1, 1, 0),
-        )
-        self.stg2_high_band_net = BaseNet(
-            nout // 4 + 2, nout // 2, self.nin_lstm // 2, nout_lstm // 2
-        )
-
-        self.stg3_full_band_net = BaseNet(
-            3 * nout // 4 + 2, nout, self.nin_lstm, nout_lstm
-        )
-
-        self.out = nn.Conv2d(nout, 2, 1, bias=False)
-        self.aux_out = nn.Conv2d(3 * nout // 4, 2, 1, bias=False)
-
-    def forward(self, x):
-        x = x[:, :, : self.max_bin]
-
-        bandw = x.size()[2] // 2
-        l1_in = x[:, :, :bandw]
-        h1_in = x[:, :, bandw:]
-        l1 = self.stg1_low_band_net(l1_in)
-        h1 = self.stg1_high_band_net(h1_in)
-        aux1 = torch.cat([l1, h1], dim=2)
-
-        l2_in = torch.cat([l1_in, l1], dim=1)
-        h2_in = torch.cat([h1_in, h1], dim=1)
-        l2 = self.stg2_low_band_net(l2_in)
-        h2 = self.stg2_high_band_net(h2_in)
-        aux2 = torch.cat([l2, h2], dim=2)
-
-        f3_in = torch.cat([x, aux1, aux2], dim=1)
-        f3 = self.stg3_full_band_net(f3_in)
-
-        mask = torch.sigmoid(self.out(f3))
-        mask = F.pad(
-            input=mask,
-            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
-            mode="replicate",
-        )
-
-        if self.training:
-            aux = torch.cat([aux1, aux2], dim=1)
-            aux = torch.sigmoid(self.aux_out(aux))
-            aux = F.pad(
-                input=aux,
-                pad=(0, 0, 0, self.output_bin - aux.size()[2]),
-                mode="replicate",
-            )
-            return mask, aux
-        else:
-            return mask
-
-    def predict_mask(self, x):
-        mask = self.forward(x)
-
-        if self.offset > 0:
-            mask = mask[:, :, :, self.offset : -self.offset]
-            assert mask.size()[3] > 0
-
-        return mask
-
-    def predict(self, x, aggressiveness=None):
-        mask = self.forward(x)
-        pred_mag = x * mask
-
-        if self.offset > 0:
-            pred_mag = pred_mag[:, :, :, self.offset : -self.offset]
-            assert pred_mag.size()[3] > 0
-
-        return pred_mag

infer/lib/uvr5_pack/lib_v5/spec_utils.py

@@ -5,8 +5,6 @@ import os
 
 import librosa
 import numpy as np
-import soundfile as sf
-from tqdm import tqdm
 
 
 def crop_center(h1, h2):
@@ -520,153 +518,3 @@ def istft(spec, hl):
     wave_left = librosa.istft(spec_left, hop_length=hl)
     wave_right = librosa.istft(spec_right, hop_length=hl)
     wave = np.asfortranarray([wave_left, wave_right])
-
-
-if __name__ == "__main__":
-    import argparse
-    import sys
-    import time
-
-    import cv2
-    from model_param_init import ModelParameters
-
-    p = argparse.ArgumentParser()
-    p.add_argument(
-        "--algorithm",
-        "-a",
-        type=str,
-        choices=["invert", "invert_p", "min_mag", "max_mag", "deep", "align"],
-        default="min_mag",
-    )
-    p.add_argument(
-        "--model_params",
-        "-m",
-        type=str,
-        default=os.path.join("modelparams", "1band_sr44100_hl512.json"),
-    )
-    p.add_argument("--output_name", "-o", type=str, default="output")
-    p.add_argument("--vocals_only", "-v", action="store_true")
-    p.add_argument("input", nargs="+")
-    args = p.parse_args()
-
-    start_time = time.time()
-
-    if args.algorithm.startswith("invert") and len(args.input) != 2:
-        raise ValueError("There should be two input files.")
-
-    if not args.algorithm.startswith("invert") and len(args.input) < 2:
-        raise ValueError("There must be at least two input files.")
-
-    wave, specs = {}, {}
-    mp = ModelParameters(args.model_params)
-
-    for i in range(len(args.input)):
-        spec = {}
-
-        for d in range(len(mp.param["band"]), 0, -1):
-            bp = mp.param["band"][d]
-
-            if d == len(mp.param["band"]):  # high-end band
-                wave[d], _ = librosa.load(
-                    args.input[i],
-                    bp["sr"],
-                    False,
-                    dtype=np.float32,
-                    res_type=bp["res_type"],
-                )
-
-                if len(wave[d].shape) == 1:  # mono to stereo
-                    wave[d] = np.array([wave[d], wave[d]])
-            else:  # lower bands
-                wave[d] = librosa.resample(
-                    wave[d + 1],
-                    mp.param["band"][d + 1]["sr"],
-                    bp["sr"],
-                    res_type=bp["res_type"],
-                )
-
-            spec[d] = wave_to_spectrogram(
-                wave[d],
-                bp["hl"],
-                bp["n_fft"],
-                mp.param["mid_side"],
-                mp.param["mid_side_b2"],
-                mp.param["reverse"],
-            )
-
-        specs[i] = combine_spectrograms(spec, mp)
-
-    del wave
-
-    if args.algorithm == "deep":
-        d_spec = np.where(np.abs(specs[0]) <= np.abs(spec[1]), specs[0], spec[1])
-        v_spec = d_spec - specs[1]
-        sf.write(
-            os.path.join("{}.wav".format(args.output_name)),
-            cmb_spectrogram_to_wave(v_spec, mp),
-            mp.param["sr"],
-        )
-
-    if args.algorithm.startswith("invert"):
-        ln = min([specs[0].shape[2], specs[1].shape[2]])
-        specs[0] = specs[0][:, :, :ln]
-        specs[1] = specs[1][:, :, :ln]
-
-        if "invert_p" == args.algorithm:
-            X_mag = np.abs(specs[0])
-            y_mag = np.abs(specs[1])
-            max_mag = np.where(X_mag >= y_mag, X_mag, y_mag)
-            v_spec = specs[1] - max_mag * np.exp(1.0j * np.angle(specs[0]))
-        else:
-            specs[1] = reduce_vocal_aggressively(specs[0], specs[1], 0.2)
-            v_spec = specs[0] - specs[1]
-
-            if not args.vocals_only:
-                X_mag = np.abs(specs[0])
-                y_mag = np.abs(specs[1])
-                v_mag = np.abs(v_spec)
-
-                X_image = spectrogram_to_image(X_mag)
-                y_image = spectrogram_to_image(y_mag)
-                v_image = spectrogram_to_image(v_mag)
-
-                cv2.imwrite("{}_X.png".format(args.output_name), X_image)
-                cv2.imwrite("{}_y.png".format(args.output_name), y_image)
-                cv2.imwrite("{}_v.png".format(args.output_name), v_image)
-
-                sf.write(
-                    "{}_X.wav".format(args.output_name),
-                    cmb_spectrogram_to_wave(specs[0], mp),
-                    mp.param["sr"],
-                )
-                sf.write(
-                    "{}_y.wav".format(args.output_name),
-                    cmb_spectrogram_to_wave(specs[1], mp),
-                    mp.param["sr"],
-                )
-
-        sf.write(
-            "{}_v.wav".format(args.output_name),
-            cmb_spectrogram_to_wave(v_spec, mp),
-            mp.param["sr"],
-        )
-    else:
-        if not args.algorithm == "deep":
-            sf.write(
-                os.path.join("ensembled", "{}.wav".format(args.output_name)),
-                cmb_spectrogram_to_wave(ensembling(args.algorithm, specs), mp),
-                mp.param["sr"],
-            )
-
-    if args.algorithm == "align":
-        trackalignment = [
-            {
-                "file1": '"{}"'.format(args.input[0]),
-                "file2": '"{}"'.format(args.input[1]),
-            }
-        ]
-
-        for i, e in tqdm(enumerate(trackalignment), desc="Performing Alignment..."):
-            os.system(f"python lib/align_tracks.py {e['file1']} {e['file2']}")
-
-    # print('Total time: {0:.{1}f}s'.format(time.time() - start_time, 1))

infer/lib/uvr5_pack/utils.py

@@ -1,17 +1,8 @@
-import json
-
 import numpy as np
 import torch
 from tqdm import tqdm
 
 
-def load_data(file_name: str = "./infer/lib/uvr5_pack/name_params.json") -> dict:
-    with open(file_name, "r") as f:
-        data = json.load(f)
-
-    return data
-
-
 def make_padding(width, cropsize, offset):
     left = offset
     roi_size = cropsize - left * 2
@@ -97,25 +88,3 @@ def inference(X_spec, device, model, aggressiveness, data):
         return (pred + pred_tta) * 0.5 * coef, X_mag, np.exp(1.0j * X_phase)
     else:
         return pred * coef, X_mag, np.exp(1.0j * X_phase)
-
-
-def _get_name_params(model_path, model_hash):
-    data = load_data()
-    flag = False
-    ModelName = model_path
-    for type in list(data):
-        for model in list(data[type][0]):
-            for i in range(len(data[type][0][model])):
-                if str(data[type][0][model][i]["hash_name"]) == model_hash:
-                    flag = True
-                elif str(data[type][0][model][i]["hash_name"]) in ModelName:
-                    flag = True
-
-                if flag:
-                    model_params_auto = data[type][0][model][i]["model_params"]
-                    param_name_auto = data[type][0][model][i]["param_name"]
-                    if type == "equivalent":
-                        return param_name_auto, model_params_auto
-                    else:
-                        flag = False
-    return param_name_auto, model_params_auto

infer/modules/uvr5/vr.py

@@ -11,7 +11,7 @@ import torch
 from infer.lib.uvr5_pack.lib_v5 import nets_61968KB as Nets
 from infer.lib.uvr5_pack.lib_v5 import spec_utils
 from infer.lib.uvr5_pack.lib_v5.model_param_init import ModelParameters
-from infer.lib.uvr5_pack.lib_v5.nets_new import CascadedNet
+from infer.lib.uvr5_pack.lib_v5.nets import CascadedNet
 from infer.lib.uvr5_pack.utils import inference