When building a CNN on tensorflow how do I specify which convolutional filter to use?

Question

I know that there are different kinds of Convolutional Filters depending on the job you want to do. ie. Sharpening, blurring, etc. Is there a specific kind we need to use for image classification?

An example CNN used for image classification is provided on the tensorflow website:

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10))

I realize the convolutional layer is using a 3x3 filter, but how do I know what type of matrix it is? A sharpening one, blurring one, etc? Can any 3X3 be used for image classification?

Answer 1

它不是 Instagram 意义上的“过滤器”。它是一个在输入图像上滑动的矩阵，乘以相应的值，然后将这些值相加。权重是可训练的，因此它们成为特征，即它们成为通过乘法提取最多意义的东西。

如果你初始化一个卷积层，它只是随机值：

import tensorflow as tf

conv = tf.keras.layers.Conv2D(filters=1, kernel_size=3)
conv.build(input_shape=(None, 28, 28, 1))

weights, biases = conv.weights

print(tf.squeeze(weights))

tf.Tensor(
[[ 0.4269786  -0.07291955  0.31100047]
 [ 0.05227929 -0.19417122 -0.12701556]
 [-0.27614036 -0.12557104 -0.12314937]], shape=(3, 3), dtype=float32)

正如我所说，过滤器是可训练值的矩阵，所以如果你的任务是检测线，你最终可能会得到这样的权重：

tf.eye(3)

<tf.Tensor: shape=(3, 3), dtype=float32, numpy=
array([[1., 0., 0.],
       [0., 1., 0.],
       [0., 0., 1.]], dtype=float32)>

下一张图像的输入将是原始图像的扭曲版本，因为它与随机浮点值相乘。

import tensorflow as tf
from skimage import data
import numpy as np
import matplotlib.pyplot as plt

conv = tf.keras.layers.Conv2D(filters=3, kernel_size=3)
image = data.chelsea()

conv.build(input_shape=(None, *image.shape))

plt.imshow(image)
plt.show()

image = np.expand_dims(image, axis=0).astype(np.float32)
for i in range(5):
    image = conv(image)

plt.imshow(np.squeeze(np.abs(np.minimum(image, 255))).astype(int))
plt.show()

第二层输入的大小也会稍微小一些，因为图像缩小了(filter_size - 1)，除非有填充。