运行ENet-SAD代码遇AttributeError：tensorflow无Summary属性

This error occurs because the code was written for TensorFlow 1.x, but Google Colab uses TensorFlow 2.x by default—where the tf.Summary API was either moved or replaced. Here are two straightforward fixes:

Solution 1: Use TF1 compatibility layer #

Update the code to reference the compatibility module for TF1-style summaries:

def scalar_summary(self, tag, value, step):
    """Log a scalar variable."""
    summary = tf.compat.v1.Summary(value=[tf.compat.v1.Summary.Value(tag=tag, simple_value=value)])
    self.writer.add_summary(summary, step)

This preserves the original logic while working with TF2.

Solution 2: Migrate to TF2's native summary API #

For a more modern approach, replace the function with TF2's built-in scalar summary method:

def scalar_summary(self, tag, value, step):
    """Log a scalar variable."""
    from tensorflow.summary import scalar
    scalar(tag, value, step=step)

Note: If your existing self.writer is a TF1-style writer, you’ll need to replace it with tf.summary.create_file_writer() to match the TF2 API.

Bonus: Full TF1 compatibility #

If the rest of the ENet-SAD code relies heavily on TF1 syntax, add this at the top of your notebook to disable TF2 behavior:

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()

This lets you use all TF1 APIs without the compat.v1 prefix.

内容的提问来源于stack exchange，提问作者Gideon Gibson