def __init__(self, in_planes, hourglass_in_planes, batch_norm=True):
super(DeepPrunerAggregator, self).__init__()
self.in_planes = in_planes
self.hourglass_in_planes = hourglass_in_planes
self.batch_norm = batch_norm
self.dres0 = nn.Sequential(
conv3d_bn_relu(batch_norm,
in_planes,
64,
kernel_size=3,
stride=1,
padding=1,
bias=False),
conv3d_bn_relu(batch_norm,
64,
32,
kernel_size=3,
stride=1,
padding=1,
bias=False),
)
self.dres1 = nn.Sequential(
conv3d_bn_relu(batch_norm,
32,
32,
kernel_size=3,
stride=1,
padding=1,
bias=False),
conv3d_bn_relu(batch_norm,
32,
hourglass_in_planes,
kernel_size=3,
stride=1,
padding=1,
bias=False),
)
self.dres2 = HWHourglass(hourglass_in_planes, batch_norm=batch_norm)
self.classify = nn.Sequential(
conv3d_bn_relu(batch_norm,
hourglass_in_planes,
hourglass_in_planes * 2,
kernel_size=3,
stride=1,
padding=1,
bias=False),
nn.Conv3d(hourglass_in_planes * 2,
1,
kernel_size=3,
stride=1,
padding=1,
bias=False))
def __init__(self, in_planes, batchNorm=True):
super(Hourglass, self).__init__()
self.batchNorm = batchNorm
self.conv1 = conv3d_bn_relu(self.batchNorm,
in_planes,
in_planes * 2,
kernel_size=3,
stride=2,
padding=1,
bias=False)
self.conv2 = conv3d_bn(self.batchNorm,
in_planes * 2,
in_planes * 2,
kernel_size=3,
stride=1,
padding=1,
bias=False)
self.conv3 = conv3d_bn_relu(self.batchNorm,
in_planes * 2,
in_planes * 2,
kernel_size=3,
stride=2,
padding=1,
bias=False)
self.conv4 = conv3d_bn_relu(self.batchNorm,
in_planes * 2,
in_planes * 2,
kernel_size=3,
stride=1,
padding=1,
bias=False)
self.conv5 = deconv3d_bn(self.batchNorm,
in_planes * 2,
in_planes * 2,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False)
self.conv6 = deconv3d_bn(self.batchNorm,
in_planes * 2,
in_planes,
kernel_size=3,
padding=1,
output_padding=1,
stride=2,
bias=False)
def _make_layer(self, in_planes, out_planes, stride=1):
return conv3d_bn_relu(self.batch_norm,
in_planes,
out_planes,
kernel_size=3,
stride=stride,
padding=1,
dilation=1,
bias=False)
def __init__(self, max_disp, in_planes=64, batch_norm=True):
super(AcfAggregator, self).__init__()
self.max_disp = max_disp
self.in_planes = in_planes
self.batch_norm = batch_norm
self.dres0 = nn.Sequential(
conv3d_bn_relu(batch_norm, self.in_planes, 32, 3, 1, 1),
conv3d_bn_relu(batch_norm, 32, 32, 3, 1, 1),
)
self.dres1 = nn.Sequential(conv3d_bn_relu(batch_norm, 32, 32, 3, 1, 1),
conv3d_bn(batch_norm, 32, 32, 3, 1, 1))
self.dres2 = Hourglass(in_planes=32, batch_norm=batch_norm)
self.dres3 = Hourglass(in_planes=32, batch_norm=batch_norm)
self.dres4 = Hourglass(in_planes=32, batch_norm=batch_norm)
self.classif1 = nn.Sequential(
conv3d_bn_relu(batch_norm, 32, 32, 3, 1, 1),
nn.Conv3d(32, 1, kernel_size=3, stride=1, padding=1, bias=False),
)
self.classif2 = nn.Sequential(
conv3d_bn_relu(batch_norm, 32, 32, 3, 1, 1),
nn.Conv3d(32, 1, kernel_size=3, stride=1, padding=1, bias=False),
)
self.classif3 = nn.Sequential(
conv3d_bn_relu(batch_norm, 32, 32, 3, 1, 1),
nn.Conv3d(32, 1, kernel_size=3, stride=1, padding=1, bias=False))
self.deconv1 = nn.ConvTranspose3d(1, 1, 8, 4, 2, bias=False)
self.deconv2 = nn.ConvTranspose3d(1, 1, 8, 4, 2, bias=False)
self.deconv3 = nn.ConvTranspose3d(1, 1, 8, 4, 2, bias=False)
def __init__(self, in_planes, hourglass_in_planes, disparity_sample_number, batch_norm=True):
super(ConfidenceRangePredictor, self).__init__()
self.in_planes = in_planes
self.hourglass_in_planes = hourglass_in_planes
self.disparity_sample_number = disparity_sample_number
self.batch_norm = batch_norm
self.dres0 = nn.Sequential(
conv3d_bn_relu(batch_norm, in_planes, 64, kernel_size=3, stride=1, padding=1, bias=False),
conv3d_bn_relu(batch_norm, 64, 32, kernel_size=3, stride=1, padding=1, bias=False),
)
self.dres1 = nn.Sequential(
conv3d_bn_relu(batch_norm, 32, 32, kernel_size=3, stride=1, padding=1, bias=False),
conv3d_bn_relu(batch_norm, 32, hourglass_in_planes, kernel_size=3, stride=1, padding=1, bias=False),
)
self.min_disparity_predictor = nn.Sequential(
HWHourglass(hourglass_in_planes, batch_norm),
conv3d_bn_relu(batch_norm, hourglass_in_planes, hourglass_in_planes*2,
kernel_size=3, stride=1, padding=1, bias=False),
nn.Conv3d(hourglass_in_planes*2, 1, kernel_size=3, stride=1, padding=1, bias=False)
)
self.max_disparity_predictor = nn.Sequential(
HWHourglass(hourglass_in_planes, batch_norm),
conv3d_bn_relu(batch_norm, hourglass_in_planes, hourglass_in_planes*2,
kernel_size=3, stride=1, padding=1, bias=False),
nn.Conv3d(hourglass_in_planes*2, 1, kernel_size=3, stride=1, padding=1, bias=False)
)
# batch norm cannot be used here, as disparity map is the input and output
self.min_disparity_conv = nn.Sequential(
nn.Conv2d(1, 1, kernel_size=5, stride=1, padding=2, bias=True),
nn.ReLU(inplace=True)
)
# batch norm cannot be used here, as disparity map is the input and output
self.max_disparity_conv = nn.Sequential(
nn.Conv2d(1, 1, kernel_size=5, stride=1, padding=2, bias=True),
nn.ReLU(inplace=True)
)
self.min_disparity_feature_conv = conv_bn_relu(batch_norm, disparity_sample_number, disparity_sample_number,
kernel_size=5, stride=1, padding=2, dilation=1, bias=True)
self.max_disparity_feature_conv = conv_bn_relu(batch_norm, disparity_sample_number, disparity_sample_number,
kernel_size=5, stride=1, padding=2, dilation=1, bias=True)
def __init__(self, max_disp, in_planes=32, batch_norm=True, num=4):
super(StereoNetAggregator, self).__init__()
self.max_disp = max_disp
self.in_planes = in_planes
self.batch_norm = batch_norm
self.num = num
self.classify = nn.ModuleList([
conv3d_bn_relu(self.batch_norm,
in_planes,
32,
kernel_size=3,
stride=1,
padding=1,
dilation=1,
bias=True) for _ in range(self.num)
])
self.lastconv = nn.Conv3d(32,
1,
kernel_size=3,
stride=1,
padding=1,
bias=True)
def __init__(self, in_planes, batch_norm=True):
super(HWHourglass, self).__init__()
self.batch_norm = batch_norm
self.conv1_a = conv3d_bn_relu(self.batch_norm,
in_planes,
in_planes * 2,
kernel_size=3,
stride=(1, 2, 2),
padding=1,
bias=False)
self.conv1_b = conv3d_bn_relu(self.batch_norm,
in_planes * 2,
in_planes * 2,
kernel_size=3,
stride=(1, 1, 1),
padding=1,
bias=False)
self.conv1_d = deconv3d_bn(self.batch_norm,
in_planes * 2,
in_planes,
kernel_size=3,
padding=1,
output_padding=(0, 1, 1),
stride=(1, 2, 2),
bias=False)
self.conv2_a = conv3d_bn_relu(self.batch_norm,
in_planes * 2,
in_planes * 4,
kernel_size=3,
stride=(1, 2, 2),
padding=1,
bias=False)
self.conv2_b = conv3d_bn_relu(self.batch_norm,
in_planes * 4,
in_planes * 4,
kernel_size=3,
stride=(1, 1, 1),
padding=1,
bias=False)
self.conv2_d = deconv3d_bn(self.batch_norm,
in_planes * 4,
in_planes * 2,
kernel_size=3,
padding=1,
output_padding=(0, 1, 1),
stride=(1, 2, 2),
bias=False)
self.conv3_a = conv3d_bn_relu(self.batch_norm,
in_planes * 4,
in_planes * 8,
kernel_size=3,
stride=(1, 2, 2),
padding=1,
bias=False)
self.conv3_b = conv3d_bn_relu(self.batch_norm,
in_planes * 8,
in_planes * 8,
kernel_size=3,
stride=(1, 1, 1),
padding=1,
bias=False)
self.conv3_d = deconv3d_bn(self.batch_norm,
in_planes * 8,
in_planes * 4,
kernel_size=3,
padding=1,
output_padding=(0, 1, 1),
stride=(1, 2, 2),
bias=False)