Question

tf.layers.conv2d和tf.layers.Conv2D有什么区别？

为什么在此paper的示例代码中使用tf.layers.Conv2D？

这是完整的代码段：

class AddCoords(base.Layer):
    """Add coords to a tensor"""
    def __init__(self, x_dim=64, y_dim=64, with_r=False):
        super(AddCoords, self).__init__()
        self.x_dim = x_dim
        self.y_dim = y_dim
        self.with_r = with_r

    def call(self, input_tensor):
        """
        input_tensor: (batch, x_dim, y_dim, c)
        """
        batch_size_tensor = tf.shape(input_tensor)[0]
        xx_ones = tf.ones([batch_size_tensor, self.x_dim], dtype=tf.int32)
        xx_ones = tf.expand_dims(xx_ones, -1)
        xx_range = tf.tile(tf.expand_dims(tf.range(self.x_dim), 0), [batch_size_tensor, 1])
        xx_range = tf.expand_dims(xx_range, 1)
        xx_channel = tf.matmul(xx_ones, xx_range)
        xx_channel = tf.expand_dims(xx_channel, -1)
        yy_ones = tf.ones([batch_size_tensor, self.y_dim], dtype=tf.int32)
        yy_ones = tf.expand_dims(yy_ones, 1)
        yy_range = tf.tile(tf.expand_dims(tf.range(self.y_dim), 0), [batch_size_tensor, 1])
        yy_range = tf.expand_dims(yy_range, -1)
        yy_channel = tf.matmul(yy_range, yy_ones)
        yy_channel = tf.expand_dims(yy_channel, -1)
        xx_channel = tf.cast(xx_channel, 'float32') / (self.x_dim - 1)
        yy_channel = tf.cast(yy_channel, 'float32') / (self.y_dim - 1)
        xx_channel = xx_channel*2 - 1
        yy_channel = yy_channel*2 - 1
        ret = tf.concat([input_tensor, xx_channel, yy_channel], axis=-1)
        if self.with_r:
            rr = tf.sqrt(tf.square(xx_channel-0.5) + tf.square(yy_channel-0.5))
            ret = tf.concat([ret, rr], axis=-1)
        return ret

class CoordConv(base.Layer):
    """CoordConv layer as in the paper."""
    def __init__(self, x_dim, y_dim, with_r, *args, **kwargs):
        super(CoordConv, self).__init__()
        self.addcoords = AddCoords(x_dim=x_dim, y_dim=y_dim, with_r=with_r)
        self.conv = tf.layers.Conv2D(*args, **kwargs)
        def call(self, input_tensor):
            ret = self.addcoords(input_tensor)
            ret = self.conv(ret)
            return ret

Answer 1

const arr1 = [ { name: "name1", id:"1" }, { name: "name2", id:"2" }, { name: "name3", id:"3" } ]; const arr2 = [ { surname: "surname1" }, { surname: "surname2" }, { surname: "surname3" } ]; const newArr = arr1.map((item, index) => ({ ...item, surname: arr2[index].surname})); console.log(newArr);是一个简单的tf.layers.conv2d / function，用于计算其输入的卷积，因此需要输入method和feature maps或kernel运行此方法。一个用户只是调用此方法来计算卷积。
但是filter是tf.layers.Conv2d（一种OOP概念），您应该在使用某些Class之前实例化它。实例化后，可以向其输入不同的输入并获取输出。使用该filter来设计和编程新的Class或一种新的操作类型，以供其他用户使用。

Answer 2

通话

y = tf.layers.conv2d(inputs, **kwargs)

现在exactly the same为

y = tf.layers.Conv2D(**kwargs).apply(inputs)

因此，只要您不打算重用由tf.layers.conv2d创建的对象而不是调用其Conv2D方法，就可以将apply视为快捷方式。

tf.layers.conv2d和tf.layers.Conv2D有什么区别？

2 个答案: