github/DeepFaceLab
1
0
Fork 0
mirror of https://github.com/iperov/DeepFaceLab.git synced 2025-08-22 14:24:40 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

Formatting

Browse source
This commit is contained in:
Jeremy Hummel 2019-08-10 09:15:16 -07:00
commit fcbc8b125c
1 changed files with 134 additions and 117 deletions
Download patch file Download diff file Expand all files Collapse all files

35
nnlib/nnlib.py
View file

@ -11,6 +11,7 @@ from utils import std_utils
from .device import device from .device import device
from interact import interact as io from interact import interact as io
class nnlib(object): class nnlib(object):
device = device # forwards nnlib.devicelib to device in order to use nnlib as standalone lib device = device # forwards nnlib.devicelib to device in order to use nnlib as standalone lib
DeviceConfig = device.Config DeviceConfig = device.Config
@ -36,7 +37,6 @@ class nnlib(object):
code_import_dlib = None code_import_dlib = None
ResNet = None ResNet = None
UNet = None UNet = None
UNetTemporalPredictor = None UNetTemporalPredictor = None
@ -115,7 +115,6 @@ UNetTemporalPredictor = nnlib.UNetTemporalPredictor
NLayerDiscriminator = nnlib.NLayerDiscriminator NLayerDiscriminator = nnlib.NLayerDiscriminator
""" """
@staticmethod @staticmethod
def _import_tf(device_config): def _import_tf(device_config):
if nnlib.tf is not None: if nnlib.tf is not None:
@ -165,7 +164,8 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
nnlib._import_tf(device_config) nnlib._import_tf(device_config)
elif nnlib.backend == "plaidML": elif nnlib.backend == "plaidML":
os.environ["KERAS_BACKEND"] = "plaidml.keras.backend" os.environ["KERAS_BACKEND"] = "plaidml.keras.backend"
os.environ["PLAIDML_DEVICE_IDS"] = ",".join ( [ nnlib.device.getDeviceID(idx) for idx in device_config.gpu_idxs] ) os.environ["PLAIDML_DEVICE_IDS"] = ",".join(
[nnlib.device.getDeviceID(idx) for idx in device_config.gpu_idxs])
# if "tensorflow" in nnlib.backend: # if "tensorflow" in nnlib.backend:
# nnlib.keras = nnlib.tf.keras # nnlib.keras = nnlib.tf.keras
@ -206,6 +206,7 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
def modelify(model_functor): def modelify(model_functor):
def func(tensor): def func(tensor):
return keras.models.Model(tensor, model_functor(tensor)) return keras.models.Model(tensor, model_functor(tensor))
return func return func
nnlib.modelify = modelify nnlib.modelify = modelify
@ -233,7 +234,9 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
outputs += [K.conv2d(inputs[i], K.constant(gauss_kernel), strides=(1, 1), padding="same")] outputs += [K.conv2d(inputs[i], K.constant(gauss_kernel), strides=(1, 1), padding="same")]
return K.concatenate(outputs, axis=-1) return K.concatenate(outputs, axis=-1)
return func return func
nnlib.gaussian_blur = gaussian_blur nnlib.gaussian_blur = gaussian_blur
def style_loss(gaussian_blur_radius=0.0, loss_weight=1.0, wnd_size=0, step_size=1): def style_loss(gaussian_blur_radius=0.0, loss_weight=1.0, wnd_size=0, step_size=1):
@ -269,13 +272,17 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
k = (sh - wnd_size) // step_size + 1 k = (sh - wnd_size) // step_size + 1
if gaussian_blur_radius > 0.0: if gaussian_blur_radius > 0.0:
target, style = gblur(target), gblur(style) target, style = gblur(target), gblur(style)
target = nnlib.tf.image.extract_image_patches(target, [1,k,k,1], [1,1,1,1], [1,step_size,step_size,1], 'VALID') target = nnlib.tf.image.extract_image_patches(target, [1, k, k, 1], [1, 1, 1, 1],
style = nnlib.tf.image.extract_image_patches(style, [1,k,k,1], [1,1,1,1], [1,step_size,step_size,1], 'VALID') [1, step_size, step_size, 1], 'VALID')
style = nnlib.tf.image.extract_image_patches(style, [1, k, k, 1], [1, 1, 1, 1],
[1, step_size, step_size, 1], 'VALID')
return sd(target, style, loss_weight) return sd(target, style, loss_weight)
if nnlib.PML is not None: if nnlib.PML is not None:
print("Sorry, plaidML backend does not support style_loss") print("Sorry, plaidML backend does not support style_loss")
return 0 return 0
return func return func
nnlib.style_loss = style_loss nnlib.style_loss = style_loss
def dssim(kernel_size=11, k1=0.01, k2=0.03, max_value=1.0): def dssim(kernel_size=11, k1=0.01, k2=0.03, max_value=1.0):
@ -461,6 +468,7 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
GAN Custom Scal Layer GAN Custom Scal Layer
Code borrows from https://github.com/flyyufelix/cnn_finetune Code borrows from https://github.com/flyyufelix/cnn_finetune
""" """
def __init__(self, weights=None, axis=-1, gamma_init='zero', **kwargs): def __init__(self, weights=None, axis=-1, gamma_init='zero', **kwargs):
self.axis = axis self.axis = axis
self.gamma_init = keras.initializers.get(gamma_init) self.gamma_init = keras.initializers.get(gamma_init)
@ -485,6 +493,7 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
config = {"axis": self.axis} config = {"axis": self.axis}
base_config = super(Scale, self).get_config() base_config = super(Scale, self).get_config()
return dict(list(base_config.items()) + list(config.items())) return dict(list(base_config.items()) + list(config.items()))
nnlib.Scale = Scale nnlib.Scale = Scale
class Adam(keras.optimizers.Optimizer): class Adam(keras.optimizers.Optimizer):
@ -590,15 +599,18 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
'amsgrad': self.amsgrad} 'amsgrad': self.amsgrad}
base_config = super(Adam, self).get_config() base_config = super(Adam, self).get_config()
return dict(list(base_config.items()) + list(config.items())) return dict(list(base_config.items()) + list(config.items()))
nnlib.Adam = Adam nnlib.Adam = Adam
def CAInitializerMP(conv_weights_list): def CAInitializerMP(conv_weights_list):
# Convolution Aware Initialization https://arxiv.org/abs/1702.06295 # Convolution Aware Initialization https://arxiv.org/abs/1702.06295
result = CAInitializerMPSubprocessor ( [ (i, K.int_shape(conv_weights)) for i, conv_weights in enumerate(conv_weights_list) ], K.floatx(), K.image_data_format() ).run() result = CAInitializerMPSubprocessor(
[(i, K.int_shape(conv_weights)) for i, conv_weights in enumerate(conv_weights_list)], K.floatx(),
K.image_data_format()).run()
for idx, weights in result: for idx, weights in result:
K.set_value(conv_weights_list[idx], weights) K.set_value(conv_weights_list[idx], weights)
nnlib.CAInitializerMP = CAInitializerMP
nnlib.CAInitializerMP = CAInitializerMP
if backend == "plaidML": if backend == "plaidML":
class TileOP_ReflectionPadding2D(nnlib.PMLTile.Operation): class TileOP_ReflectionPadding2D(nnlib.PMLTile.Operation):
@ -644,7 +656,8 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
raise NotImplemented raise NotImplemented
super(TileOP_ReflectionPadding2D, self).__init__(c, [('I', input)], super(TileOP_ReflectionPadding2D, self).__init__(c, [('I', input)],
[('O', nnlib.PMLTile.Shape(input.shape.dtype, out_dims ) )]) [('O', nnlib.PMLTile.Shape(input.shape.dtype,
out_dims))])
class ReflectionPadding2D(keras.layers.Layer): class ReflectionPadding2D(keras.layers.Layer):
def __init__(self, padding=(1, 1), **kwargs): def __init__(self, padding=(1, 1), **kwargs):
@ -697,6 +710,7 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
if self.reflect_pad: if self.reflect_pad:
x = ReflectionPadding2D(self.pad)(x) x = ReflectionPadding2D(self.pad)(x)
return self.func(x) return self.func(x)
nnlib.Conv2D = Conv2D nnlib.Conv2D = Conv2D
class Conv2DTranspose(): class Conv2DTranspose():
@ -717,6 +731,7 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
if self.reflect_pad: if self.reflect_pad:
x = ReflectionPadding2D(self.pad)(x) x = ReflectionPadding2D(self.pad)(x)
return self.func(x) return self.func(x)
nnlib.Conv2DTranspose = Conv2DTranspose nnlib.Conv2DTranspose = Conv2DTranspose
@staticmethod @staticmethod
@ -771,6 +786,7 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
def __init__(self, mask, is_mse=False): def __init__(self, mask, is_mse=False):
self.mask = mask self.mask = mask
self.is_mse = is_mse self.is_mse = is_mse
def __call__(self, y_true, y_pred): def __call__(self, y_true, y_pred):
total_loss = None total_loss = None
mask = self.mask mask = self.mask
@ -779,8 +795,8 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
return K.mean(50 * K.square(y_true * blur_mask - y_pred * blur_mask)) return K.mean(50 * K.square(y_true * blur_mask - y_pred * blur_mask))
else: else:
return 10 * dssim()(y_true * mask, y_pred * mask) return 10 * dssim()(y_true * mask, y_pred * mask)
nnlib.DSSIMMSEMaskLoss = DSSIMMSEMaskLoss
nnlib.DSSIMMSEMaskLoss = DSSIMMSEMaskLoss
''' '''
def ResNet(output_nc, use_batch_norm, ngf=64, n_blocks=6, use_dropout=False): def ResNet(output_nc, use_batch_norm, ngf=64, n_blocks=6, use_dropout=False):
@ -967,6 +983,7 @@ NLayerDiscriminator = nnlib.NLayerDiscriminator
return func return func
nnlib.NLayerDiscriminator = NLayerDiscriminator nnlib.NLayerDiscriminator = NLayerDiscriminator
''' '''
@staticmethod @staticmethod
def finalize_all(): def finalize_all():
if nnlib.keras_contrib is not None: if nnlib.keras_contrib is not None: