mnist_cond_lsgan.py 文件源码

def generator(input_var,Y):
    yb = Y#.dimshuffle(0, 1, 'x', 'x')
    G_1 = lasagne.layers.InputLayer(shape=(None, NLAT,1,1),input_var=input_var)

    G_2 = lasagne.layers.InputLayer(shape=(None, 10),input_var=Y)
    network=G_1
    network_yb=G_2

    network = conv_layer(network, 1, 4 * 4 * 128, 1, 'valid')
    #print(input_var.shape[0])
    network = ll.ReshapeLayer(network, (-1, 128, 4, 4))

    network = CondConvConcatLayer([network,network_yb])
    network = resnet_block(network, 3,138)
    network = CondConvConcatLayer([network,network_yb])

    #network = resnet_block(network, 3, 128)
    network = BilinearUpsampling(network, ratio=2)
    network = batch_norm(conv_layer(network, 3,138, 1, 'same'))
    network = CondConvConcatLayer([network,network_yb])

    network = resnet_block(network, 3, 148)
    network = CondConvConcatLayer([network,network_yb])

    network = BilinearUpsampling(network, ratio=2)
    network = batch_norm(conv_layer(network, 3, 32, 1, 'valid'))
    network = BilinearUpsampling(network, ratio=2)
    network = batch_norm(conv_layer(network, 3, 32, 1, 'same'))
    network = CondConvConcatLayer([network,network_yb])

    #network =  resnet_block(network, 3, 32)
    network = conv_layer(network, 1, 1, 1, 'valid', nonlinearity=sigmoid)
    #network =lasagne.layers.Conv2DLayer(network, num_filters=1, filter_size=1, stride=1, nonlinearity=sigmoid)
    return network, G_1,G_2

# In[23]: