def __init__(self, filters, sn=False, use_bias=False, **kwargs): super(rk_2D, self).__init__() self.l1_conv2d = BaseLayers.Conv2D(filters=filters, kernel_size=[3, 3], strides=[1, 1], padding="REFLECT", use_bias=use_bias, sn=sn, **kwargs) self.l2_norm = BaseLayers.InstanceNorm2D() self.l3_activation = BaseLayers.ReLU() self.l4_conv2d = BaseLayers.Conv2D(filters=filters, kernel_size=[3, 3], strides=[1, 1], padding="REFLECT", use_bias=use_bias, sn=sn, **kwargs) self.l5_norm = BaseLayers.InstanceNorm2D()
def __init__(self, filters, sn=False, **kwargs): super(ckd_5s2_2D, self).__init__() self.filters_zp = filters self.l1_conv2d = BaseLayers.Conv2D(filters=filters, kernel_size=[5, 5], strides=[2, 2], padding='SAME', sn=sn, **kwargs) self.l2_activation = BaseLayers.LeakyReLU() self.l3_dropout = BaseLayers.Dropout(rate=0.3)
def __init__(self,filters,sn=False,use_bias=False,activation="relu",**kwargs): super(c7s1_k_2D,self).__init__() self.filters_zp = filters self.l1_conv2d = BaseLayers.Conv2D(filters=filters,kernel_size=[7,7],strides=[1,1],padding="REFLECT",use_bias=use_bias,sn=sn,**kwargs) self.l2_norm = BaseLayers.InstanceNorm2D() if activation.lower()=="relu": self.l3_activation = BaseLayers.ReLU() elif activation.lower()=="sigmoid": self.l3_activation = tf.keras.layers.Activation("sigmoid") elif activation.lower()=="tanh": self.l3_activation = tf.keras.layers.Activation("tanh") else: raise ValueError("Un supported activation "+activation)
def __init__(self, use_sigmoid=True, sn=False, **kwargs): super(last_conv_2D, self).__init__() self.l1_conv2d = BaseLayers.Conv2D(filters=1, kernel_size=[4, 4], strides=[1, 1], padding='SAME', use_bias=True, sn=sn, **kwargs) if use_sigmoid: self.l2_activation = tf.keras.layers.Activation("sigmoid") else: self.l2_activation = tf.keras.layers.Activation("linear")
def __init__(self, filters, sn=False, norm=True, use_bias=False, **kwargs): super(ck_2D, self).__init__() self.filters_zp = filters self.l1_conv2d = BaseLayers.Conv2D(filters=filters, kernel_size=[4, 4], strides=[2, 2], padding='SAME', use_bias=use_bias, sn=sn, **kwargs) if norm: self.l2_norm = BaseLayers.InstanceNorm2D() else: self.l2_norm = tf.keras.layers.Activation("linear") self.l3_activation = BaseLayers.LeakyReLU(alpha=0.2)
tf.config.experimental.set_memory_growth(physical_devices[0], True) from tensorflow.keras.mixed_precision import experimental as mixed_precision policy = mixed_precision.Policy('mixed_float16') mixed_precision.set_policy(policy) # a = c7s1_k_2D(6,sn=True,use_bias=True) # a.build(input_shape=[None,256,256,3]) # w = tf.random.normal([1,256,256,3]) # import time # for __ in range(5): # start = time.time() # for _ in range(1000): # y = a(w) # print(time.time()-start) l1_conv2d = BaseLayers.Conv2D(filters=64,kernel_size=[7,7],strides=[1,1],padding="REFLECT",use_bias=False,sn=False) # l1_conv2d = tf.keras.layers.Conv2D(filters=64,kernel_size=[7,7],strides=[1,1],padding="VALID",use_bias=False) l1_conv2d.build(input_shape=[None,256,256,8]) w = tf.random.normal([8,256,256,8]) import time for __ in range(5): start = time.time() for _ in range(1000): y = l1_conv2d(w) print(time.time()-start) # l1_conv2d = tf.keras.layers.Dense(4096) # l1_conv2d.build(input_shape=[None,784]) # w = tf.random.normal([4096,784]) # import time # for __ in range(5):