收敛最快的魔改ESPCN, DepthwiseConv2D

原创

wmlgl

工作日志

2023/02/25 17:58

阅读数 126

本文被收录于专区

AI & 大数据

进入专区参与更多专题讨论

3 月 22 日，南京，聊聊生成式 AI 应用构建

    input_image = Input(shape=(None, None, self.channels), name='x')
    x = DepthwiseConv2D( 3, depth_multiplier= int(64 / self.channels), kernel_initializer="he_normal",
               padding='same', activation="tanh", name="conv1")(input_image)
    x = DepthwiseConv2D(5, kernel_initializer="he_normal",
               padding='same', activation="relu", name="conv2")(x)
    x = Conv2D(self.scale_factor**2*self.channels, 3, kernel_initializer="he_normal",
               padding='same', activation="relu", name="conv3")(x)
    y = tf.nn.depth_to_space(x, self.scale_factor, name="shuffle")
    model = Model(inputs=input_image, outputs=y)
    model.summary()

展开阅读全文