net = caffe.NetSpec()
net.data = L.MemoryData(dim=[1, 1], ntop=1)
net.label = L.MemoryData(dim=[1, 1], ntop=1, include=[dict(phase=0)])
net.sknet = ML.SKNet(net.data,
ip_depth=0,
dropout=0,
fmap_inc_rule = lambda fmaps: 80,
fmap_dec_rule = lambda fmaps: 80,
fmap_bridge_rule = lambda fmaps: 3,
fmap_start=80,
conv=[[3,3,3],[3,3,3],[3,3,3],[3,3,3],[3,3,3],[3,3,3]],
pool=[[2,2,2],[2,2,2],[2,2,2],[1,1,1]],
padding=[85,85,85])
net.prob = L.Softmax(net.sknet, ntop=1, in_place=False, include=[dict(phase=1)])
net.loss = L.SoftmaxWithLoss(net.sknet, net.label, ntop=0, loss_weight=1.0, include=[dict(phase=0)])
pygt.fix_input_dims(net,
[net.data, net.label],
max_shapes = [[130,130,130],[130,130,130]],
shape_coupled = [-1, 0, 0])
protonet = net.to_proto()
protonet.name = 'net';
# Store the network as prototxt
with open(protonet.name + '.prototxt', 'w') as f:
print(protonet, file=f)
from caffe import params as P
from caffe import to_proto
from pygreentea.pygreentea import metalayers as ML
net = caffe.NetSpec()
net.data = L.MemoryData(dim=[1, 1], ntop=1)
net.label = L.MemoryData(dim=[1, 1], ntop=1, include=[dict(phase=0)])
net.sknet = ML.SKNet(net.data,
fmap_start=64,
conv=[[1,8,8],[2,6,6],[2,4,4]],
pool=[[1,2,2],[1,2,2],[1,2,2]],
padding=[8,88,88])
net.out = L.Convolution(net.sknet, kernel_size=[1,1,1], num_output=9, weight_filler=dict(type='msra'), bias_filler=dict(type='constant'))
net.prob = L.Softmax(net.out, ntop=1, in_place=False, include=[dict(phase=1)])
net.loss = L.SoftmaxWithLoss(net.out, net.label, ntop=0, loss_weight=1.0, include=[dict(phase=0)])
pygt.fix_input_dims(net,
[net.data, net.label],
max_shapes = [[9,250,250],[9,250,250]],
shape_coupled = [-1, -1, 1])
protonet = net.to_proto()
protonet.name = 'net';
# Store the network as prototxt
with open(protonet.name + '.prototxt', 'w') as f:
print(protonet, file=f)
net.aff_pred = L.Sigmoid(net.aff_out, ntop=1, in_place=False)
net.smax_pred = L.Softmax(net.smax_out, ntop=1, in_place=False, include=[dict(phase=1)])
# Choose a loss function and input data, label and scale inputs. Only include it during the training phase (phase = 0)
net.euclid_loss = L.EuclideanLoss(net.aff_pred, net.aff_label, net.scale, ntop=0, loss_weight=10.0, include=[dict(phase=0, stage='euclid')])
net.malis_loss = L.MalisLoss(net.aff_pred, net.aff_label, net.comp_label, net.nhood, ntop=0, loss_weight=5.0, include=[dict(phase=0, stage='malis')])
net.smax_loss = L.SoftmaxWithLoss(net.smax_out, net.smax_label, ntop=0, loss_weight=1.0, include=[dict(phase=0)])
# Fix the spatial input dimensions. Note that only spatial dimensions get modified, the minibatch size
# and the channels/feature maps must be set correctly by the user (since this code can definitely not
# figure out the user's intent). If the code does not seem to terminate, then the issue is most likely
# a wrong number of feature maps / channels in either the MemoryData-layers or the network output.
# This function takes as input:
# - The network
# - A list of other inputs to test (note: the nhood input is static and not spatially testable, thus excluded here)
# - A list of the maximal shapes for each input
# - A list of spatial dependencies; here [-1, 0] means the Y axis is a free parameter, and the X axis should be identical to the Y axis.
pygt.fix_input_dims(net,
[net.data, net.aff_label, net.comp_label, net.smax_label, net.scale],
max_shapes = [[500,500],[500,500],[500,500],[500,500],[500,500]],
shape_coupled = [-1, 0])
protonet = net.to_proto()
protonet.name = 'net';
# Store the network as prototxt
with open(protonet.name + '.prototxt', 'w') as f:
print(protonet, file=f)
