def forward(self, x, width_mult=1):
module = self.body[0]
res = common.SlimModule(x, module, width_mult)
res = self.act(res)
module = self.body[1]
res = common.SlimModule(res, module, width_mult)
res = self.caLayer(res, width_mult)
res += x
return res
def forward(self, x, width_mult=1):
y = self.avg_pool(x)
module = getattr(self.conv_du, '0')
y = common.SlimModule(y, module, width_mult)
y = self.relu(y)
module = getattr(self.conv_du, '1')
y = common.SlimModule(y, module, width_mult)
y = self.sigmoid(y)
return x * y
def forward(self, x, width_mult=1):
x = self.sub_mean(x)
weight = self.head_conv.weight
n_feats = weight.shape[0]
out_ch = int(n_feats * width_mult)
weight = weight[:out_ch, :self.n_colors, :, :]
bias = self.head_conv.bias[:out_ch]
x = nn.functional.conv2d(x,
weight,
bias,
stride=self.head_conv.stride,
padding=self.head_conv.padding)
res = x
for module in self.body:
res = module(res, width_mult)
res = common.SlimModule(res, self.body_conv, width_mult)
res += x
x = self.upsampler(res, width_mult)
weight = self.tail_conv.weight[:self.n_colors, :out_ch, :, :]
bias = self.tail_conv.bias[:self.n_colors]
x = nn.functional.conv2d(x,
weight,
bias,
stride=self.tail_conv.stride,
padding=self.tail_conv.padding)
x = self.add_mean(x)
return x
def forward(self, x, width_mult):
res = x
for module in self.body:
res = module(res, width_mult)
res = common.SlimModule(res, self.conv, width_mult)
res += x
return res
def forward(self, x, width_mult):
out = common.SlimModule(x, self.conv, width_mult)
out = self.act(out)
return out