def __init__(self, inplanes):
super(hourglass, self).__init__()
self.conv1 = nn.Sequential(
submodule.convbn_3d(inplanes,
inplanes * 2,
kernel_size=3,
stride=2,
pad=1), nn.ReLU(inplace=True))
self.conv2 = submodule.convbn_3d(inplanes * 2,
inplanes * 2,
kernel_size=3,
stride=1,
pad=1)
self.conv3 = nn.Sequential(
submodule.convbn_3d(inplanes * 2,
inplanes * 2,
kernel_size=3,
stride=2,
pad=1), nn.ReLU(inplace=True))
self.conv4 = nn.Sequential(
submodule.convbn_3d(inplanes * 2,
inplanes * 2,
kernel_size=3,
stride=1,
pad=1), nn.ReLU(inplace=True))
self.conv5 = nn.Sequential(
nn.ConvTranspose3d(inplanes * 2,
inplanes * 2,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False),
nn.BatchNorm3d(inplanes * 2)) #+conv2
self.conv6 = nn.Sequential(
nn.ConvTranspose3d(inplanes * 2,
inplanes,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(inplanes)) #+x
def __init__(self, maxdisp):
super(PSMNet, self).__init__()
self.maxdisp = maxdisp
self.feature_extraction = submodule.feature_extraction()
self.dres0 = nn.Sequential(submodule.convbn_3d(64, 32, 3, 1, 1),
nn.ReLU(inplace=True),
submodule.convbn_3d(32, 32, 3, 1, 1),
nn.ReLU(inplace=True))
self.dres1 = nn.Sequential(submodule.convbn_3d(32, 32, 3, 1, 1),
nn.ReLU(inplace=True),
submodule.convbn_3d(32, 32, 3, 1, 1))
self.dres2 = hourglass(32)
self.dres3 = hourglass(32)
self.dres4 = hourglass(32)
self.classif1 = nn.Sequential(
submodule.convbn_3d(32, 32, 3, 1, 1), nn.ReLU(inplace=True),
nn.Conv3d(32, 1, kernel_size=3, padding=1, stride=1, bias=False))
self.classif2 = nn.Sequential(
submodule.convbn_3d(32, 32, 3, 1, 1), nn.ReLU(inplace=True),
nn.Conv3d(32, 1, kernel_size=3, padding=1, stride=1, bias=False))
self.classif3 = nn.Sequential(
submodule.convbn_3d(32, 32, 3, 1, 1), nn.ReLU(inplace=True),
nn.Conv3d(32, 1, kernel_size=3, padding=1, stride=1, bias=False))
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.Conv3d):
n = m.kernel_size[0] * m.kernel_size[1] * m.kernel_size[
2] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm3d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
def __init__(self, maxdisp):
super(PSMNet, self).__init__()
self.maxdisp = maxdisp
self.inplanes = 32
self.feature_extraction = sb.feature_extraction()
self.dres0 = nn.Sequential(sb.convbn_3d(64, 32, 3, 1, 1),
nn.ReLU(inplace=True))
self.dres1 = nn.Sequential(sb.convbn_3d(32, 32, 3, 1, 1),
nn.ReLU(inplace=True),
sb.convbn_3d(32, 32, 3, 1, 1))
self.dres2 = nn.Sequential(
sb.convbn_3d(self.inplanes,
self.inplanes,
kernel_size=3,
stride=2,
pad=1), nn.ReLU(inplace=True),
sb.convbn_3d(self.inplanes,
self.inplanes,
kernel_size=3,
stride=1,
pad=1))
self.dres3 = nn.Sequential(
sb.convbn_3d(self.inplanes,
self.inplanes * 2,
kernel_size=3,
stride=2,
pad=1), nn.ReLU(inplace=True),
sb.convbn_3d(self.inplanes * 2,
self.inplanes * 2,
kernel_size=3,
stride=1,
pad=1))
self.dres4 = nn.Sequential(
sb.convbn_3d(self.inplanes * 2,
self.inplanes * 2,
kernel_size=3,
stride=2,
pad=1), nn.ReLU(inplace=True),
sb.convbn_3d(self.inplanes * 2,
self.inplanes * 2,
kernel_size=3,
stride=1,
pad=1))
self.classif1 = nn.Sequential(sb.convbn_3d(128, 64, 3, 1, 1),
nn.ReLU(inplace=True),
nn.Conv3d(64, 32, 3, 1, 1))
self.classif2 = nn.Sequential(sb.convbn_3d(128, 64, 3, 1, 1),
nn.ReLU(inplace=True),
nn.Conv3d(64, 32, 3, 1, 1))
self.conv1_1 = nn.Sequential(sb.convbn_3d(32, 32, 3, 1, 1),
nn.ReLU(inplace=True))
self.conv2_1 = nn.Sequential(sb.convbn_3d(32, 32, 3, 1, 1),
nn.ReLU(inplace=True))
self.dres5 = nn.Sequential(
sb.convbn_3d(self.inplanes * 2,
self.inplanes,
kernel_size=3,
stride=1,
pad=1), nn.ReLU(inplace=True))
self.deconv3_1 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes * 2,
self.inplanes * 2,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes * 2))
self.deconv4_1 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes * 2,
self.inplanes,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes))
self.deconv5_1 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes,
self.inplanes,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes))
self.deconv6_1 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes,
self.inplanes,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes))
self.deconv3_2 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes * 2,
self.inplanes * 2,
kernel_size=5,
padding=2,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes * 2))
self.deconv4_2 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes * 2,
self.inplanes,
kernel_size=5,
padding=2,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes))
self.deconv5_2 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes,
self.inplanes,
kernel_size=5,
padding=2,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes))
self.deconv6_2 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes,
self.inplanes,
kernel_size=5,
padding=2,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes))
self.deconv1 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes,
self.inplanes,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(self.inplanes))
self.deconv2 = nn.Sequential(
nn.ConvTranspose3d(self.inplanes,
1,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False), nn.BatchNorm3d(1))
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.Conv3d):
n = m.kernel_size[0] * m.kernel_size[1] * m.kernel_size[
2] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm3d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()