TensorFlow CNN报错：输出深度无法被组数整除，如何解决？

问题背景 #

基于Kaggle机器视觉课程实现CNN时，运行代码抛出错误，仅能显示原始车辆图像和卷积核，后续无输出。谷歌搜索后未找到匹配的解决方案。

原始代码 #

import numpy as np
from itertools import product

import tensorflow as tf
import matplotlib.pyplot as plt
import sys

import keras
from keras import layers


def show_kernel(kernel, label=True, digits=None, text_size=28):
    # Format kernel
    kernel = np.array(kernel)
    if digits is not None:
        kernel = kernel.round(digits)

    # Plot kernel
    cmap = plt.get_cmap('Blues_r')
    plt.imshow(kernel, cmap=cmap)
    rows, cols = kernel.shape
    thresh = (kernel.max() + kernel.min())/2
    # Optionally, add value labels
    if label:
        for i, j in product(range(rows), range(cols)):
            val = kernel[i, j]
            color = cmap(0) if val > thresh else cmap(255)
            plt.text(j, i, val,
                     color=color, size=text_size,
                     horizontalalignment='center', verticalalignment='center')
    plt.xticks([])
    plt.yticks([])

plt.rc('figure', autolayout=True)
plt.rc('axes', labelweight='bold', labelsize='large',
       titleweight='bold', titlesize=18, titlepad=10)
plt.rc('image', cmap='magma')

img_source = sys.path[0] + "/car_feature.png"

image_path = img_source
image = tf.io.read_file(image_path)
image = tf.io.decode_png(image)

plt.figure(figsize=(6, 6))
plt.imshow(tf.squeeze(image), cmap='gray')
plt.axis('off')


kernel = tf.constant([
    [-1, -1, -1],
    [-1,  8, -1],
    [-1, -1, -1],
])

plt.figure(figsize=(3, 3))

# Reformat for batch compatibility.
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
image = tf.expand_dims(image, axis=0)
kernel = tf.reshape(kernel, [*kernel.shape, 1, 1])
kernel = tf.cast(kernel, dtype=tf.float32)

image_filter = tf.nn.conv2d(
    input=image,
    filters=kernel,
    strides=1,
    padding='SAME',
)

plt.figure(figsize=(6, 6))
plt.imshow(tf.squeeze(image_filter))
plt.axis('off')
plt.show();

报错信息 #

Traceback (most recent call last):
  File "c:\Users\q621208\WorkSpace\Tensorflow\Keras\cnn_relu.py", line 64, in <module>
    image_filter = tf.nn.conv2d(
  File "C:\Users\q621208\AppData\Local\Programs\Python\Python310\lib\site-packages\tensorflow\python\util\traceback_utils.py", line 153, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "C:\Users\q621208\AppData\Local\Programs\Python\Python310\lib\site-packages\tensorflow\python\framework\ops.py", line 7262, in raise_from_not_ok_status
    raise core._status_to_exception(e) from None  # pylint: disable=protected-access
tensorflow.python.framework.errors_impl.InvalidArgumentError: {{function_node __wrapped__Conv2D_device_/job:localhost/replica:0/task:0/device:CPU:0}} output depth must be evenly divisible by number of groups: 1 vs 4 [Op:Conv2D]

期望效果 #

得到经过卷积核处理后的车辆图像（边缘增强效果），原始图像为彩色车辆图，处理后图像应突出车辆轮廓。

错误原因 #

tf.io.decode_png读取的PNG图像默认是4通道（RGBA，包含透明度通道），而定义的卷积核是单通道（适配灰度图），导致卷积操作时输入通道数（4）与核的输入通道数（1）不匹配，触发组数不兼容的错误。

解决方法 #

有两种可行方案：

方案1：将图像转为单通道灰度图 #

在读取图像后添加一行代码，把4通道图像转成灰度图：

image = tf.io.decode_png(image)
# 添加此行
image = tf.image.rgb_to_grayscale(image)

方案2：调整卷积核适配4通道输入 #

如果需要保留彩色图像，将单通道卷积核复制4份，匹配输入的4通道：

kernel = tf.reshape(kernel, [*kernel.shape, 1, 1])
kernel = tf.cast(kernel, dtype=tf.float32)
# 添加此行，将核复制4份适配4通道
kernel = tf.tile(kernel, [1,1,4,1])

修改后的完整代码（方案1示例） #

import numpy as np
from itertools import product

import tensorflow as tf
import matplotlib.pyplot as plt
import sys

import keras
from keras import layers


def show_kernel(kernel, label=True, digits=None, text_size=28):
    # Format kernel
    kernel = np.array(kernel)
    if digits is not None:
        kernel = kernel.round(digits)

    # Plot kernel
    cmap = plt.get_cmap('Blues_r')
    plt.imshow(kernel, cmap=cmap)
    rows, cols = kernel.shape
    thresh = (kernel.max() + kernel.min())/2
    # Optionally, add value labels
    if label:
        for i, j in product(range(rows), range(cols)):
            val = kernel[i, j]
            color = cmap(0) if val > thresh else cmap(255)
            plt.text(j, i, val,
                     color=color, size=text_size,
                     horizontalalignment='center', verticalalignment='center')
    plt.xticks([])
    plt.yticks([])

plt.rc('figure', autolayout=True)
plt.rc('axes', labelweight='bold', labelsize='large',
       titleweight='bold', titlesize=18, titlepad=10)
plt.rc('image', cmap='magma')

img_source = sys.path[0] + "/car_feature.png"

image_path = img_source
image = tf.io.read_file(image_path)
image = tf.io.decode_png(image)
# 转为灰度图
image = tf.image.rgb_to_grayscale(image)

plt.figure(figsize=(6, 6))
plt.imshow(tf.squeeze(image), cmap='gray')
plt.axis('off')


kernel = tf.constant([
    [-1, -1, -1],
    [-1,  8, -1],
    [-1, -1, -1],
])

plt.figure(figsize=(3, 3))
show_kernel(kernel)  # 显示卷积核

# Reformat for batch compatibility.
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
image = tf.expand_dims(image, axis=0)
kernel = tf.reshape(kernel, [*kernel.shape, 1, 1])
kernel = tf.cast(kernel, dtype=tf.float32)

image_filter = tf.nn.conv2d(
    input=image,
    filters=kernel,
    strides=1,
    padding='SAME',
)

plt.figure(figsize=(6, 6))
plt.imshow(tf.squeeze(image_filter), cmap='gray')
plt.axis('off')
plt.show();

内容的提问来源于stack exchange，提问作者Ratmir