执行图像分类Python代码时遇Conv2D报错：输入非符号张量

Hey there, let's work through this error you're hitting when setting up your CNN's convolution layer. This issue almost always ties back to version mismatches or API inconsistencies, so here are the most reliable fixes:

1. Switch to TensorFlow's Integrated Keras API (Most Probable Fix) #

If you're using the standalone keras library with a recent TensorFlow version, they're likely conflicting. Modern TensorFlow includes its own fully integrated Keras implementation (tf.keras) that eliminates these tensor-related bugs.

Update your imports to use tf.keras instead:

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense

# Initialising the CNN
classifier = Sequential()
# Step 1 - Convolution
classifier.add(Conv2D(32, (3, 3), input_shape=(64, 64, 3), activation='relu'))

2. Verify Keras/TensorFlow Version Compatibility #

If you want to stick with the standalone Keras library, make sure it’s compatible with your installed TensorFlow version. For reference:

• Keras 2.3.x pairs with TensorFlow 1.15.x
• Keras 2.4+ is built for TensorFlow 2.x

Check your current versions with these commands:

pip show keras tensorflow

If they’re mismatched, upgrade or downgrade to compatible releases.

3. Double-Check Input Data Format (For Future Training) #

While your error happens during layer setup, if you run into follow-up issues after fixing imports, confirm your training data matches the input_shape=(64,64,3) specification. Your data should be a 4D tensor with the shape (number_of_samples, 64, 64, 3) (batch size, height, width, color channels).

内容的提问来源于stack exchange，提问作者Naveen Nair