def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.x = gan.session.run(gan.inputs.x) batch = self.x.shape[0] self.x = np.reshape( self.x[0], [1, self.x.shape[1], self.x.shape[2], self.x.shape[3]]) self.x = np.tile(self.x, [batch, 1, 1, 1])
def __init__(self, gan, samples_per_row=4, session=None): BaseSampler.__init__(self, gan, samples_per_row) self.latent1 = self.gan.latent.next() self.latent2 = self.gan.latent.next() self.velocity = 2 / 30.0 direction = self.gan.latent.next() self.origin = direction self.pos = self.latent1 self.hardtanh = nn.Hardtanh() g_params = self.gan.latent_parameters() if self.latent1.shape[1] // 2 == g_params[0].shape[1]: #recombine a split g_params = [ torch.cat([p1, p2], 1) for p1, p2 in zip(g_params[:len(g_params) // 2], g_params[len(g_params) // 2:]) ] self.eigvec = torch.svd(torch.cat(g_params, 0)).V #self.eigvec = torch.svd(list(self.gan.g_parameters())[0]).V self.index = 0 self.direction = self.eigvec[:, self.index].unsqueeze(0) self.direction = self.direction / torch.norm(self.direction) self.ones = torch.ones_like(self.direction, device="cuda:0") self.mask = torch.cat([ torch.zeros([1, direction.shape[1] // 2]), torch.ones([1, direction.shape[1] // 2]) ], dim=1).cuda() self.mask = torch.ones_like(self.mask).cuda() self.steps = 30
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) x_t = gan.inputs.x global x x = gan.session.run(x_t) self.samplers = [] default = tf.zeros_like(gan.generator.sample) def add_samples(layer): layer = gan.generator.layer(layer) if layer is None: layer = default self.samplers.append( IdentitySampler( gan, tf.image.resize_images(layer, [256, 256], method=1), 1)) add_samples('g8x8') add_samples('g16x16') add_samples('g32x32') add_samples('g64x64') add_samples('g128x128') add_samples('g256x256')
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.z = None self.y = None self.x = None self.step = 0 self.steps = 30 self.target = None
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.z = None self.y = None self.x = None self.g_t = self.replace_none(gan.generator.sample) self.rows = 4 self.columns = 8
def __init__(self, gan, samples_per_row=8): sess = gan.session self.x = gan.session.run(gan.preview) print("__________", np.shape(self.x),'---oo') frames = np.shape(self.x)[1]//height self.frames=frames self.x = np.split(self.x, frames, axis=1) self.i = 0 BaseSampler.__init__(self, gan, samples_per_row)
def __init__(self, gan, samples_per_row=8, session=None): BaseSampler.__init__(self, gan, samples_per_row) self.z_start = None self.y = None self.x = None self.step = 0 self.steps = [] self.step_count = 30 self.target = None
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) x_t = gan.inputs.x global x x = gan.session.run(x_t) self.samplers = [ #IdentitySampler(gan, gan.inputs.x, samples_per_row), #IdentitySampler(gan, gan.inputs.xb, samples_per_row), #IdentitySampler(gan, gan.autoencoded_x, samples_per_row), #StaticBatchSampler(gan, samples_per_row), #BatchSampler(gan, samples_per_row), #RandomWalkSampler(gan, samples_per_row) ] #self.samplers += [IdentitySampler(gan, tf.image.resize_images(gan.inputs.x, [128,128], method=1), samples_per_row)] if hasattr(gan.generator, 'pe_layers'): self.samplers += [ IdentitySampler(gan, gx, samples_per_row) for gx in gan.generator.pe_layers ] pe_layers = self.gan.skip_connections.get_array( "progressive_enhancement") #self.samplers += if hasattr(gan, 'autoencoded_x'): self.samplers += [ IdentitySampler( gan, tf.concat([gan.inputs.x, gan.autoencoded_x], axis=0), samples_per_row) ] if gan.config.loss['class'] == BoundaryEquilibriumLoss: self.samplers += [BeganSampler(gan, samples_per_row)] if isinstance(gan.generator, SegmentGenerator): self.samplers += [SegmentSampler(gan)] if hasattr(gan, 'seq'): self.samplers += [ IdentitySampler( gan, tf.image.resize_images(gx, [128, 128], method=1), samples_per_row) for gx in gan.seq ] default = gan.generator.sample #tf.zeros_like(gan.generator.layer('gend8x8')) def add_samples(layer): layer = gan.generator.layer(layer) if layer is None: layer = default self.samplers.append( IdentitySampler( gan, tf.image.resize_images(layer, [128, 128], method=1), 1)) add_samples('gend8x8') add_samples('gend16x16')
def __init__(self, gan, samples_per_row=8, session=None): BaseSampler.__init__(self, gan, samples_per_row) self.z = None self.y = None self.x = None self.step = 0 self.steps = 30 self.target = None self.z_t = gan.uniform_distribution.sample self.z_v = gan.session.run(self.z_t) self.styleb_t = gan.styleb.sample
def __init__(self, gan, samples_per_row=8, session=None): BaseSampler.__init__(self, gan, samples_per_row) self.z_start = None self.y = None self.x = None self.step = 0 self.steps = [] self.step_count = 30 self.target = None self.rows = 2 self.columns = 4 self.needed = int(self.rows * self.columns / gan.batch_size())
def __init__(self, gan, samples_per_row=8, session=None): BaseSampler.__init__(self, gan, samples_per_row) self.z_start = None self.y = None self.x = None self.step = 0 self.steps = [] self.step_count = 30 self.target = None self.rows = 2 self.columns = 4 self.needed = int(self.rows*self.columns / gan.batch_size())
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.latent = self.gan.latent.next().data.clone() self.x = torch.cat([ torch.unsqueeze(self.gan.x[0], 0).repeat(gan.batch_size() // 2, 1, 1, 1), torch.unsqueeze(self.gan.x[1], 0).repeat(gan.batch_size() // 2, 1, 1, 1) ], 0) self.bw = self.x.mean(axis=1, keepdims=True).repeat(1, 3, 1, 1).to(gan.device) self.gan = gan
def __init__(self, gan, samples_per_row=4, session=None): BaseSampler.__init__(self, gan, samples_per_row) self.latent1 = self.gan.latent.next() self.latent2 = self.gan.latent.next() self.velocity = 15 / 24.0 direction = self.gan.latent.next() self.origin = direction self.pos = self.latent1 self.direction = direction / torch.norm( direction, p=2, dim=1, keepdim=True).expand_as(direction) self.hardtanh = nn.Hardtanh() self.ones = torch.ones_like(self.direction, device="cuda:0") self.mask = torch.cat([ torch.zeros([1, direction.shape[1] // 2]), torch.ones([1, direction.shape[1] // 2]) ], dim=1).cuda() self.mask = torch.ones_like(self.mask).cuda()
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) x_t = gan.inputs.x global x x = gan.session.run(x_t) self.samplers = [ #IdentitySampler(gan, gan.inputs.x, samples_per_row), #IdentitySampler(gan, gan.inputs.xb, samples_per_row), #IdentitySampler(gan, gan.autoencoded_x, samples_per_row), #StaticBatchSampler(gan, samples_per_row), #BatchSampler(gan, samples_per_row), #RandomWalkSampler(gan, samples_per_row) ] #self.samplers += [IdentitySampler(gan, tf.image.resize_images(gan.inputs.x, [128,128], method=1), samples_per_row)] if hasattr(gan.generator, 'pe_layers'): self.samplers += [IdentitySampler(gan, gx, samples_per_row) for gx in gan.generator.pe_layers] pe_layers = self.gan.skip_connections.get_array("progressive_enhancement") #self.samplers += if hasattr(gan, 'autoencoded_x'): self.samplers += [IdentitySampler(gan, tf.concat([gan.inputs.x,gan.autoencoded_x], axis=0), samples_per_row)] if gan.config.loss['class'] == BoundaryEquilibriumLoss: self.samplers += [BeganSampler(gan, samples_per_row)] if isinstance(gan.generator, SegmentGenerator): self.samplers += [SegmentSampler(gan)] if hasattr(gan, 'seq'): self.samplers += [IdentitySampler(gan, tf.image.resize_images(gx, [128,128], method=1), samples_per_row) for gx in gan.seq] default = gan.generator.sample#tf.zeros_like(gan.generator.layer('gend8x8')) def add_samples(layer): layer = gan.generator.layer(layer) if layer is None: layer = default self.samplers.append(IdentitySampler(gan, tf.image.resize_images(layer, [128,128], method=1), 1)) add_samples('gend8x8') add_samples('gend16x16')
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.latent = self.gan.latent.next().data.clone() #self.x = torch.unsqueeze(self.gan.x[0],0).repeat(gan.batch_size(),1,1,1) self.x = self.gan.x self.bw = self.x.mean(axis=1, keepdims=True).repeat(1, 3, 1, 1).to(gan.device) self.gan = gan self.latent1 = self.gan.latent.next() self.latent2 = self.gan.latent.next() self.velocity = 15 / 24.0 direction = self.gan.latent.next() self.origin = direction self.pos = self.latent1 self.direction = direction / torch.norm( direction, p=2, dim=1, keepdim=True).expand_as(direction) self.hardtanh = nn.Hardtanh() self.ones = torch.ones_like(self.direction, device="cuda:0") self.xstep = 0 self.xstep_count = 1200
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) x_t = gan.inputs.x global x x = gan.session.run(x_t) self.samplers = [] default = tf.zeros_like(gan.generator.sample) def add_samples(layer): layer = gan.generator.layer(layer) if layer is None: layer = default self.samplers.append(IdentitySampler(gan, tf.image.resize_images(layer, [256,256], method=1), 1)) add_samples('g8x8') add_samples('g16x16') add_samples('g32x32') add_samples('g64x64') add_samples('g128x128') add_samples('g256x256')
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.z = None self.y = None self.x = None self.mask = None self.step = 0 self.steps = 8 self.target = None self.y_t = gan.y.sample self.y = gan.session.run(self.y_t) self.g=tf.get_default_graph() self.frames = gan.session.run(gan.frames) self.frames_t = gan.frames self.zs2, self.cs2 = gan.session.run([gan.zs[-1], gan.cs[-1]]) self.zs2 = [self.zs2] self.cs2 = [self.cs2] self.zs_t = [gan.video_generator_last_z] self.cs_t = [gan.video_generator_last_c] self.zs = gan.session.run([gan.video_generator_last_z]) self.cs = gan.session.run([gan.video_generator_last_c]) self.i=0
def __init__(self, gan, node, samples_per_row=8, x=None, z=None): self.node = node BaseSampler.__init__(self, gan, samples_per_row) self.z = None self.x = None
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row)
def __init__(self, gan, node, samples_per_row=8): self.node = node BaseSampler.__init__(self, gan, samples_per_row)
def __init__(self, gan): BaseSampler.__init__(self, gan) self.xs = None self.samples = 3
def __init__(self, gan): BaseSampler.__init__(self, gan) self.x_v = None self.z_v = None self.created = False self.mask_t = None
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.z = None self.y = None self.x = None self.d_real = None
def __init__(self, gan): BaseSampler.__init__(self, gan) self.xs = None self.samples = 10 self.display_count = 5
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) x_t = gan.inputs.x global x x = gan.session.run(x_t) self.samplers = [ #IdentitySampler(gan, gan.inputs.x, samples_per_row), #IdentitySampler(gan, gan.inputs.xb, samples_per_row), #IdentitySampler(gan, gan.autoencoded_x, samples_per_row), #StaticBatchSampler(gan, samples_per_row), #BatchSampler(gan, samples_per_row), #RandomWalkSampler(gan, samples_per_row) ] #self.samplers += [IdentitySampler(gan, tf.image.resize_images(gan.inputs.x, [128,128], method=1), samples_per_row)] #if hasattr(gan.generator, 'pe_layers'): # self.samplers += [IdentitySampler(gan, gx, samples_per_row) for gx in gan.generator.pe_layers] # pe_layers = self.gan.skip_connections.get_array("progressive_enhancement") if hasattr(gan, 'noise_generator'): self.samplers += [ IdentitySampler( gan, tf.concat([ gan.noise_generator.sample, gan.generator.sample, gan.noise_generator.sample + gan.generator.sample ], axis=0), samples_per_row) ] #self.samplers += if hasattr(gan, 'autoencoded_x'): self.samplers += [ IdentitySampler( gan, tf.concat([gan.inputs.x, gan.autoencoded_x], axis=0), samples_per_row) ] if gan.config.loss['class'] == BoundaryEquilibriumLoss: self.samplers += [BeganSampler(gan, samples_per_row)] if isinstance(gan.generator, SegmentGenerator): self.samplers += [SegmentSampler(gan)] if hasattr(gan, 'seq'): self.samplers += [ IdentitySampler( gan, tf.image.resize_images(gx, [128, 128], method=1), samples_per_row) for gx in gan.seq ] for train_hook in self.gan.train_hooks(): if isinstance(train_hook, RollingMemoryTrainHook): if "is_cross_replica_context" in dir( tf.distribute ) and tf.distribute.is_cross_replica_context(): for debug in train_hook.distributed_debug(): self.samplers += [ IdentitySampler(gan, debug, samples_per_row) ] else: self.samplers += [ IdentitySampler(gan, train_hook.mx, samples_per_row) ] self.samplers += [ IdentitySampler(gan, train_hook.mg, samples_per_row) ] if isinstance(train_hook, RollingMemory2TrainHook): if "is_cross_replica_context" in dir( tf.distribute ) and tf.distribute.is_cross_replica_context(): for debug in train_hook.distributed_debug(): self.samplers += [ IdentitySampler(gan, debug, samples_per_row) ] else: for v in train_hook.variables(): self.samplers += [ IdentitySampler(gan, v, samples_per_row) ] default = gan.generator.sample #tf.zeros_like(gan.generator.layer('gend8x8')) def add_samples(layer): layer = gan.generator.layer(layer) if layer is None: layer = default self.samplers.append( IdentitySampler( gan, tf.image.resize_images(layer, [128, 128], method=1), 1)) #add_samples('gend8x8') #add_samples('gend16x16') #add_samples('gend32x32') #add_samples('gend64x64') #add_samples('gend128x128') if hasattr(gan.discriminator, 'named_layers' ) and "match_support_mx" in gan.discriminator.named_layers: self.samplers.append( IdentitySampler( gan, tf.concat([ gan.inputs.x, tf.image.resize_images( gan.discriminator.named_layers['match_support_mx'], [128, 128], method=1), tf.image.resize_images( gan.discriminator. named_layers['match_support_m+x'], [128, 128], method=1) ], axis=0), 1)) self.samplers.append( IdentitySampler( gan, tf.concat([ gan.generator.sample, tf.image.resize_images( gan.discriminator.named_layers['match_support_mg'], [128, 128], method=1), tf.image.resize_images( gan.discriminator. named_layers['match_support_m+g'], [128, 128], method=1) ], axis=0), 1))
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row=samples_per_row) self.xs = None self.samples = 1
def __init__(self, gan): BaseSampler.__init__(self, gan) self.xa_v = None self.xb_v = None self.created = False
def __init__(self, gan, samples_per_row=8): self.z = None self.i = 0 BaseSampler.__init__(self, gan, samples_per_row)
def __init__(self, gan, samples_per_row): BaseSampler.__init__(self, gan, samples_per_row) self.x_v = None self.z_v = None self.created = False
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.z = None self.y = None self.x = None
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.inputs = self.gan.inputs.next() self.z = self.gan.latent.next()
def __init__(self, gan, samples_per_row=8): BaseSampler.__init__(self, gan, samples_per_row) self.latent = self.gan.latent.next().data.clone()