def __init__(self, inplanes, outplanes, batch_norm, leaky_relu):
nn.Module.__init__(self)
self.batch_norm = batch_norm
self.leaky_relu = leaky_relu
self.conv1 = scn.SubmanifoldConvolution(dimension=3,
nIn = inplanes,
nOut = outplanes,
filter_size = 3,
bias=False)
if self.batch_norm:
if self.leaky_relu: self.bn1 = scn.BatchNormLeakyReLU(outplanes)
else: self.bn1 = scn.BatchNormReLU(outplanes)
self.conv2 = scn.SubmanifoldConvolution(dimension=3,
nIn = outplanes,
nOut = outplanes,
filter_size = 3,
bias = False)
if self.batch_norm:
self.bn2 = scn.BatchNormalization(outplanes)
self.residual = scn.Identity()
if self.leaky_relu: self.relu = scn.LeakyReLU()
else: self.relu = scn.ReLU()
self.add = scn.AddTable()
def block(self,
m,
a,
b,
dimension=3,
residual_blocks=False,
leakiness=0,
kernel_size=3,
use_batch_norm=True): # default using residual_block
if use_batch_norm:
Activation = lambda channels: scn.BatchNormLeakyReLU(
channels, leakiness=leakiness)
else:
Activation = lambda channels: scn.LeakyReLU(leakiness)
if residual_blocks: #ResNet style blocks
m.add(scn.ConcatTable().add(scn.Identity(
) if a == b else scn.NetworkInNetwork(a, b, False)).add(
scn.Sequential().add(Activation(a)).add(
scn.SubmanifoldConvolution(dimension, a, b, kernel_size,
False)).add(Activation(b)).add(
scn.SubmanifoldConvolution(
dimension, b, b,
kernel_size,
False)))).add(
scn.AddTable())
else: #VGG style blocks
m.add(scn.Sequential().add(Activation(a)).add(
scn.SubmanifoldConvolution(dimension, a, b, kernel_size,
False)))
def __init__(self,
dimension,
reps,
n_layers,
leakiness=0,
input_layer=None,
name='encoder',
device=None):
super(Encoder, self).__init__()
self.dimension = dimension
self.reps = reps
self.n_layers = n_layers
self.leakiness = leakiness
self.name = name
self.device = device
if input_layer != None:
self.input_layer = scn.InputLayer(len(input_layer), input_layer)
self.blocks = []
self.block_names = {}
n_in, n_out = 1, 1
for i in range(len(n_layers)):
block = scn.Sequential()
# add reps Resnet blocks, where reps >= 1 and first block just ensures number of
# input channels is correct
for rep in range(reps):
block.add(
resnet_block(dimension,
n_in,
n_out,
1,
leakiness,
computation='submanifoldconvolution'))
n_in = n_out
n_out = n_layers[i][1]
'''
block.add(
scn.BatchNormLeakyReLU(n_in, leakiness)
)
'''
block.add(scn.LeakyReLU(leakiness))
if len(n_layers[i]) == 2:
block.add(scn.Convolution(dimension, n_in, n_out, 2, 2, False))
elif len(n_layers[i]
) == 3 and n_layers[i][2] == 'submanifoldconvolution':
block.add(
scn.SubmanifoldConvolution(dimension, n_in, n_out, 2,
False))
elif len(n_layers[i]) == 3 and n_layers[i][2] == 'maxpool':
block.add(scn.MaxPooling(dimension, 2, 2))
elif len(n_layers[i]) == 3 and n_layers[i][2] == 'avgpool':
block.add(scn.AveragePooling(dimension, 2, 2))
block_name = get_block_name(name, dimension, reps, n_in, n_out,
leakiness)
n_in = n_out
self.blocks.append(block)
self.block_names[block_name] = len(self.blocks) - 1
self.blocks = torch.nn.ModuleList(self.blocks)
def __init__(self):
super(Model, self).__init__()
self.inputLayer = scn.InputLayer(dimension, spatial_size=512, mode=3)
self.initialconv = scn.SubmanifoldConvolution(dimension, nPlanes, 64,
7, False)
self.residual = scn.Identity()
self.add = scn.AddTable()
self.sparsebl11 = scn.Sequential().add(
scn.SubmanifoldConvolution(dimension, 64, 64, 3, False)).add(
scn.BatchNormLeakyReLU(64)).add(
scn.SubmanifoldConvolution(dimension, 64, 64, 3, False))
self.sparsebl12 = scn.Sequential().add(
scn.SubmanifoldConvolution(dimension, 64, 64, 3, False)).add(
scn.BatchNormLeakyReLU(64)).add(
scn.SubmanifoldConvolution(dimension, 64, 64, 3, False))
self.sparsebl21 = scn.Sequential().add(
scn.SubmanifoldConvolution(dimension, 128, 128, 3, False)).add(
scn.BatchNormLeakyReLU(128)).add(
scn.SubmanifoldConvolution(dimension, 128, 128, 3, False))
self.sparsebl22 = scn.Sequential().add(
scn.SubmanifoldConvolution(dimension, 128, 128, 3, False)).add(
scn.BatchNormLeakyReLU(128)).add(
scn.SubmanifoldConvolution(dimension, 128, 128, 3, False))
self.relu1 = scn.LeakyReLU(64)
self.relu2 = scn.LeakyReLU(128)
self.downsample1 = scn.Sequential().add(
scn.Convolution(dimension, 64, 64, [2, 2, 2], [2, 2, 2],
False)).add(scn.BatchNormLeakyReLU(64))
self.downsample2 = scn.Sequential().add(
scn.Convolution(dimension, 64, 128, [2, 2, 2], [2, 2, 2],
False)).add(scn.BatchNormLeakyReLU(128))
self.downsample3 = scn.Sequential().add(
scn.Convolution(dimension, 128, 64, [4, 4, 4], [4, 4, 4],
False)).add(scn.BatchNormLeakyReLU(64))
self.downsample4 = scn.Sequential().add(
scn.Convolution(dimension, 64, 2, [4, 4, 4], [4, 4, 4],
False)).add(scn.BatchNormLeakyReLU(2))
self.sparsetodense = scn.SparseToDense(dimension, 2)
self.dropout1 = nn.Dropout(0.5)
self.dropout2 = nn.Dropout(0.5)
self.linear2 = nn.Linear(2 * 8 * 8 * 8, 2)
self.linear3 = nn.Linear(2, 1)
def resnet_block(dimension,
n_in,
n_out,
kernel,
leakiness=0,
computation='convolution'):
'''Build and return ResNet block
'''
assert computation in [
'submanifoldconvolution', 'convolution', 'fullconvolution',
'deconvolution'
]
if computation == 'convolution':
computation = lambda n_in, n_out: scn.Convolution(
dimension, n_in, n_out, kernel[0], kernel[1], False)
elif computation == 'submanifoldconvolution':
assert type(
kernel
) == int, f"`kernel` must be int, {type(kernel)} was provided"
computation = lambda n_in, n_out: scn.SubmanifoldConvolution(
dimension, n_in, n_out, kernel, False)
elif computation == 'deconvolution':
assert type(
kernel
) == int, f"`kernel` must be int, {type(kernel)} was provided"
computation = lambda n_in, n_out: scn.Deconvolution(
dimension, n_in, n_out, kernel, kernel, False)
else:
computation = lambda n_in, n_out: scn.FullConvolution(
dimension, n_in, n_out, kernel[0], kernel[1], False)
block = scn.Sequential()
block.add(scn.ConcatTable(
).add(scn.NetworkInNetwork(n_in, n_out, False)).add(scn.Sequential().add(
# scn.BatchNormLeakyReLU(n_in, leakiness=leakiness)
scn.LeakyReLU(leakiness)).add(computation(n_in, n_out)).add(
# scn.BatchNormLeakyReLU(n_out, leakiness=leakiness)
scn.LeakyReLU(leakiness)).add(computation(n_out, n_out)))).add(
scn.AddTable())
return block
def block(self,
m,
a,
b,
dimension=3,
residual_blocks=False,
leakiness=0,
kernel_size=3): # default using residual_block
if residual_blocks: #ResNet style blocks
m.add(scn.ConcatTable().add(scn.Identity(
) if a == b else scn.NetworkInNetwork(a, b, False)).add(
scn.Sequential().add(scn.LeakyReLU(leakiness)).add(
scn.SubmanifoldConvolution(
dimension, a, b, kernel_size,
False)).add(scn.LeakyReLU(leakiness)).add(
scn.SubmanifoldConvolution(
dimension, b, b, kernel_size,
False)))).add(scn.AddTable())
else: #VGG style blocks
m.add(scn.Sequential().add(scn.LeakyReLU(leakiness)).add(
scn.SubmanifoldConvolution(dimension, a, b, kernel_size,
False)))
def __init__(self, inplanes, outplanes, batch_norm, leaky_relu, nplanes=1):
nn.Module.__init__(self)
self.batch_norm = batch_norm
self.leaky_relu = leaky_relu
self.conv = scn.Convolution(dimension=3,
nIn=inplanes,
nOut=outplanes,
filter_size=[nplanes, 2, 2],
filter_stride=[1, 2, 2],
bias=False)
if self.batch_norm:
self.bn = scn.BatchNormalization(outplanes)
if self.leaky_relu: self.relu = scn.LeakyReLU()
else: self.relu = scn.ReLU()
def __init__(self, inplanes, outplanes, batch_norm, leaky_relu, nplanes=1):
nn.Module.__init__(self)
self.batch_norm = batch_norm
self.leaky_relu = leaky_relu
self.conv1 = scn.SubmanifoldConvolution(dimension=3,
nIn=inplanes,
nOut=outplanes,
filter_size=[nplanes, 3, 3],
bias=False)
if self.batch_norm:
if self.leaky_relu: self.bn1 = scn.BatchNormLeakyReLU(outplanes)
else: self.bn1 = scn.BatchNormReLU(outplanes)
else:
if self.leaky_relu: self.relu = scn.LeakyReLU()
else: self.relu = scn.ReLU()
def __init__(self,
dimension,
n_layers,
unet=False,
name='decoder',
use_sparsify=True,
device=None):
super(Decoder, self).__init__()
self.dimension = dimension
self.n_layers = n_layers
self.unet = unet
self.name = name
self.device = device
self.blocks = []
self.block_names = {}
for i in range(len(n_layers)):
n_in, n_out = n_layers[i][1], n_layers[i][0]
block = scn.Sequential()
'''
block.add(
scn.BatchNormLeakyReLU(n_in, 0)
)
'''
block.add(scn.LeakyReLU(0))
if len(n_layers[i]) == 2:
block.add(
scn.FullConvolution(dimension, n_in, n_out, 2, 2, False))
elif len(n_layers[i]) == 3 and n_layers[i][2] == False:
# don't upsample
block.add(
scn.SubmanifoldConvolution(dimension, n_in, n_out, 1,
False))
if use_sparsify:
block.add(scn.Sparsify(dimension, n_out))
block_name = get_block_name(name, dimension, 0, n_in, n_out, 0)
self.blocks.append(block)
self.block_names[block_name] = len(self.blocks) - 1
self.blocks = torch.nn.ModuleList(self.blocks)
def __init__(
self,
output_shape,
use_norm=True,
name='Pyramid_LightNoBN',
use_residual=True,
blocks=[
(64, 15), (80, 11), (96, 7), (128, 5)
], # define blocks, the tuple represent (num_filters, kernel). (blocks are divided by one downsample op)
layers_per_block=2, # each layer consists of 2 sscnn if use residual is true, and 1 if false
downsample_type='max_pool2',
leakiness=0,
# dense_blocks=[(160, (3, 3, 3), (2, 1, 1)), (192, (3, 3, 3), (2, 1, 1)), (224, (3, 3, 3), (2, 1, 1))], # define final dense blocks, with (num_filters, kernel, stride)
# out_filters=512,
# final_z_dim=12,
**kwargs):
super(Pyramid_LightNoBN, self).__init__()
self.name = name
self.use_residual = use_residual
self.layers_per_block = layers_per_block
self.blocks = blocks
if downsample_type == 'max_pool2':
Downsample = change_default_args(dimension=3,
pool_size=(2, 2, 2),
pool_stride=(2, 2,
2))(scn.MaxPooling)
else:
# scn.Convolution(dimension, num_filter_fpn[k-1], num_filter_fpn[k], 3, 2, False)
raise ValueError('Invalid downsample type')
sparse_shape = np.array(output_shape[1:4]) # + [1, 0, 0]
self.scn_input = scn.InputLayer(3, sparse_shape.tolist())
self.voxel_output_shape = output_shape
m = scn.Sequential()
(num_filters, kernel_size) = blocks[0]
# TODO fix, should add a subsparseconv
self.block(m,
1,
num_filters,
dimension=3,
residual_blocks=use_residual,
kernel_size=kernel_size)
for _ in range(layers_per_block - 1):
self.block(m,
num_filters,
num_filters,
dimension=3,
residual_blocks=use_residual,
kernel_size=kernel_size)
self.block_models = [m]
prev_num_filters = num_filters
for k, (num_filters, kernel_size) in enumerate(blocks[1:]):
k = k + 1
m = scn.Sequential()
# downsample
m.add(scn.LeakyReLU(leakiness))
m.add(Downsample())
self.block(m,
prev_num_filters,
num_filters,
dimension=3,
residual_blocks=use_residual,
kernel_size=kernel_size)
for _ in range(layers_per_block - 1):
self.block(m,
num_filters,
num_filters,
dimension=3,
residual_blocks=use_residual,
kernel_size=kernel_size)
self.block_models.append(m)
prev_num_filters = num_filters
self.block_models.append(scn.Sequential().add(
scn.LeakyReLU(leakiness)).add(Downsample()))
self.block_models = ListModule(*self.block_models)
self.sparse_to_dense = scn.SparseToDense(3, prev_num_filters)
self.z_combiner = nn.Sequential(
nn.Conv3d(prev_num_filters,
prev_num_filters, (8, 1, 1),
groups=prev_num_filters),
# nn.BatchNorm3d(num_filters),
nn.LeakyReLU(negative_slope=leakiness))
