def def_bfgs(model_G, layer='conv4', npx=64, alpha=0.002):
print('COMPILING...')
t = time()
# 符号化定义
x_f = T.tensor4()
x = T.tensor4()
z = T.matrix() # 随机种子
tanh = activations.Tanh()
gx = model_G(tanh(z)) # 生成的图像
if layer is 'hog':
gx_f = HOGNet.get_hog(gx, use_bin=True, BS=4)
else:
# 调整图像格式
gx_t = AlexNet.transform_im(gx)
gx_net = AlexNet.build_model(gx_t,
layer=layer,
shape=(None, 3, npx, npx))
AlexNet.load_model(gx_net, layer=layer)
# AlexNet截止在layer的输出
gx_f = lasagne.layers.get_output(gx_net[layer], deterministic=True)
f_rec = T.mean(T.sqr(x_f - gx_f), axis=(1, 2, 3)) * sharedX(alpha)
x_rec = T.mean(T.sqr(x - gx), axis=(1, 2, 3))
cost = T.sum(f_rec) + T.sum(x_rec)
grad = T.grad(cost, z)
output = [cost, grad, gx]
_invert = theano.function(inputs=[z, x, x_f], outputs=output)
print('%.2f seconds to compile _bfgs function' % (time() - t))
return _invert, z
def def_bfgs(net, layer='conv4', npx=64, alpha=0.002):
print('COMPILING...')
t = time()
x_f = T.tensor4()
x = T.tensor4()
z = T.matrix()
z = theano.printing.Print('this is z')(z)
tanh = activations.Tanh()
tz = tanh(z)
# tz = printing_op(tz)
# tz = z_scale * tz
net.labels_var = T.TensorType('float32', [False] * 512) ('labels_var')
gx = net.G.eval(z, net.labels_var, ignore_unused_inputs=True)
# gx = printing_op(gx)
# gx = misc.adjust_dynamic_range(gx, [-1,1], [0,1])
scale_factor = 16
gx = theano.tensor.signal.pool.pool_2d(gx, ds=(scale_factor, scale_factor), mode='average_exc_pad', ignore_border=True)
# gx = printing_op(gx)
if layer is 'hog':
gx_f = HOGNet.get_hog(gx, use_bin=True, BS=4)
else:
gx_t = AlexNet.transform_im(gx)
gx_net = AlexNet.build_model(gx_t, layer=layer, shape=(None, 3, npx, npx))
AlexNet.load_model(gx_net, layer=layer)
gx_f = lasagne.layers.get_output(gx_net[layer], deterministic=True)
f_rec = T.mean(T.sqr(x_f - gx_f), axis=(1, 2, 3)) * sharedX(alpha)
x_rec = T.mean(T.sqr(x - gx), axis=(1, 2, 3))
cost = T.sum(f_rec) + T.sum(x_rec)
grad = T.grad(cost, z)
output = [cost, grad, gx]
_invert = theano.function(inputs=[z, x, x_f], outputs=output)
print('%.2f seconds to compile _bfgs function' % (time() - t))
return _invert,z
dg2 = gain_ifn((ndf * 2), 'dg2')
db2 = bias_ifn((ndf * 2), 'db2')
dw3 = difn((ndf * 4, ndf * 2, 5, 5), 'dw3')
dg3 = gain_ifn((ndf * 4), 'dg3')
db3 = bias_ifn((ndf * 4), 'db3')
dw4 = difn((ndf * 8, ndf * 4, 5, 5), 'dw4')
dg4 = gain_ifn((ndf * 8), 'dg4')
db4 = bias_ifn((ndf * 8), 'db4')
dwy = difn((ndf * 8 * 4 * 4, 1), 'dwy')
dwy1 = difn((ndf * 8 * 4 * 4, 1), 'dwy')
# models
relu = activations.Rectify()
sigmoid = activations.Sigmoid()
lrelu = activations.LeakyRectify()
tanh = activations.Tanh()
bce = T.nnet.binary_crossentropy
# generator model
gen_params = [gw, gg, gb, gw2, gg2, gb2, gw3, gg3, gb3, gw4, gg4, gb4, gwx]
def gen(Z, w, g, b, w2, g2, b2, w3, g3, b3, w4, g4, b4, wx):
h = relu(batchnorm(T.dot(Z, w), g=g, b=b))
h = h.reshape((h.shape[0], ngf * 8, 4, 4))
h2 = relu(
batchnorm(deconv(h, w2, subsample=(2, 2), border_mode=(2, 2)),
g=g2,
b=b2))
h3 = relu(
batchnorm(deconv(h2, w3, subsample=(2, 2), border_mode=(2, 2)),
def mnistGANcond():
"""
This example loads the 32x32 imagenet model used in the paper,
generates 400 random samples, and sorts them according to the
discriminator's probability of being real and renders them to
the file samples.png
"""
nc = 1
npx = 28
ngf = 64 # # of gen filters in first conv layer
ndf = 128
ny = 10 # # of classes
nz = 100 # # of dim for Z
k = 1 # # of discrim updates for each gen update
l2 = 2.5e-5 # l2 weight decay
b1 = 0.5 # momentum term of adam
nc = 1 # # of channels in image
ny = 10 # # of classes
nbatch = 128 # # of examples in batch
npx = 28 # # of pixels width/height of images
nz = 100 # # of dim for Z
ngfc = 1024 # # of gen units for fully connected layers
ndfc = 1024 # # of discrim units for fully connected layers
ngf = 64 # # of gen filters in first conv layer
ndf = 64 # # of discrim filters in first conv layer
nx = npx * npx * nc # # of dimensions in X
niter = 100 # # of iter at starting learning rate
niter_decay = 100 # # of iter to linearly decay learning rate to zero
lr = 0.0002
relu = activations.Rectify()
sigmoid = activations.Sigmoid()
lrelu = activations.LeakyRectify()
tanh = activations.Tanh()
model_path = 'dcgan_code-master/mnist/models/cond_dcgan/'
gen_params = [
sharedX(p) for p in joblib.load(model_path + '200_gen_params.jl')
]
discrim_params = [
sharedX(p) for p in joblib.load(model_path + '200_discrim_params.jl')
]
def gen(Z, Y, w, w2, w3, wx):
yb = Y.dimshuffle(0, 1, 'x', 'x')
Z = T.concatenate([Z, Y], axis=1)
h = relu(batchnorm(T.dot(Z, w)))
h = T.concatenate([h, Y], axis=1)
h2 = relu(batchnorm(T.dot(h, w2)))
h2 = h2.reshape((h2.shape[0], ngf * 2, 7, 7))
h2 = conv_cond_concat(h2, yb)
h3 = relu(
batchnorm(deconv(h2, w3, subsample=(2, 2), border_mode=(2, 2))))
h3 = conv_cond_concat(h3, yb)
x = sigmoid(deconv(h3, wx, subsample=(2, 2), border_mode=(2, 2)))
return x
def discrim(X, Y, w, w2, w3, wy):
yb = Y.dimshuffle(0, 1, 'x', 'x')
X = conv_cond_concat(X, yb)
h = lrelu(dnn_conv(X, w, subsample=(2, 2), border_mode=(2, 2)))
h = conv_cond_concat(h, yb)
h2 = lrelu(
batchnorm(dnn_conv(h, w2, subsample=(2, 2), border_mode=(2, 2))))
h2 = T.flatten(h2, 2)
h2 = T.concatenate([h2, Y], axis=1)
h3 = lrelu(batchnorm(T.dot(h2, w3)))
h3 = T.concatenate([h3, Y], axis=1)
y = sigmoid(T.dot(h3, wy))
return y
def inverse_transform(X):
X = (X.reshape(-1, nc, npx, npx).transpose(0, 2, 3, 1) + 1.) / 2.
return X
Z = T.matrix()
X = T.tensor4()
Y = T.matrix()
gX = gen(Z, Y, *gen_params)
dX = discrim(X, Y, *discrim_params)
_gen = theano.function([Z, Y], gX)
_discrim = theano.function([X, Y], dX)
sample_zmb = floatX(np_rng.uniform(-1., 1., size=(200, nz)))
sample_ymb = floatX(
OneHot(
np.asarray([[i for _ in range(20)] for i in range(10)]).flatten(),
ny))
samples = _gen(sample_zmb, sample_ymb)
scores = _discrim(samples, sample_ymb)
print(scores[1:10])
sort = np.argsort(scores.flatten())[::-1]
samples = samples[sort]
print(np.shape(inverse_transform(samples)))
print(min(scores))
print(max(scores))
color_grid_vis(inverse_transform(samples), (20, 20), 'samples.png')
return inverse_transform(samples), sample_ymb
