Introduction

This TF-Slim error usually means a TensorFlow API expected a scalar or tensor of integer type, but your code passed a Python list or protobuf-style structure instead. In older TensorFlow and TF-Slim code, this often happens around shape definitions, queue configuration, or preprocessing utilities that expect plain integers. The fix is to trace the exact argument type and convert it into the tensor or integer form the API actually wants.

What The Error Is Telling You

The important part of the message is not just Expected int32, but the phrase saying a list contains tensors of type _Message. That usually points to one of these situations:

• a shape argument was built from the wrong object,
• a parser returned structured metadata that got passed into a tensor API,
• a TF-Slim helper expected scalar dimensions but received a container of tensors.

In other words, this is usually an argument-construction bug, not a math bug.

Start With A Minimal Shape Example

TensorFlow shape-sensitive APIs are strict.

1import tensorflow as tf
2
3height = tf.constant(224, dtype=tf.int32)
4width = tf.constant(224, dtype=tf.int32)
5
6shape = tf.stack([height, width])
7print(shape)

This is valid because the API receives a real tensor. Problems begin when you pass mixed Python objects that look like dimensions but are not actual integers or tensors in the expected format.

A Typical Wrong Pattern

Older TF-Slim input pipelines often build arguments from metadata objects. If you pass that metadata directly into a reshape, resize, or preprocessing function, TensorFlow may reject it.

Conceptually wrong pattern:

# Pseudocode shape idea, not valid:
# resize_shape = [some_proto_value, some_tensor]
# image = tf.image.resize(image, resize_shape)

The right fix is to extract primitive values first.

Convert Inputs Explicitly

When in doubt, normalize every shape component into tf.int32.

1import tensorflow as tf
2
3def as_int32_pair(h, w):
4    h = tf.cast(h, tf.int32)
5    w = tf.cast(w, tf.int32)
6    return tf.stack([h, w])
7
8image = tf.random.uniform((32, 32, 3))
9target_shape = as_int32_pair(64, 64)
10resized = tf.image.resize(image, target_shape)
11print(resized.shape)

This removes ambiguity about type and shape.

Debug The Actual Argument Types

In TF-Slim codebases, print or inspect the type of every input before the failing call.

print(type(suspect_value))
print(suspect_value)

If the object is a list of message wrappers, extract the numeric field first. If it is a Python list of tensors, convert it deliberately with tf.stack or tf.convert_to_tensor instead of passing the list raw.

TF-Slim Code Often Hides The Real Source

Because TF-Slim wraps many low-level TensorFlow calls, the stack trace may point at a helper rather than the actual bad argument. Read upward from the failing line and locate the exact variable being fed into the op.

Common places to inspect:

• image preprocessing size arguments,
• shape lists used for reshape,
• queue or batch-size settings,
• dataset parser outputs.

Modernization Helps

TF-Slim was useful, but many codebases are easier to maintain if moved toward tf.keras and modern tf.data pipelines. That does not magically eliminate type bugs, but it reduces some of the old wrapper complexity that makes these errors harder to read.

If the project is still active, consider treating repeated TF-Slim type issues as a signal to modernize interfaces around the failing path.

Common Pitfalls

• Passing Python lists of mixed objects into shape-sensitive TensorFlow ops.
• Assuming protobuf or config wrapper values are already plain integers.
• Debugging the failing operation without inspecting the exact argument types.
• Using TF-Slim helper layers without checking what low-level TensorFlow API they call.
• Mixing eager tensors, Python ints, and metadata objects in one shape expression.

Summary

• This error usually means an API expected integer shape data and received a structured list instead.
• Trace the exact offending argument instead of treating it as a generic TF-Slim failure.
• Convert dimension inputs explicitly to tf.int32 tensors or Python integers.
• Use tf.stack or tf.convert_to_tensor when you really need tensor-shaped lists.
• Repeated type problems in TF-Slim code are often a sign that the interface needs cleanup or modernization.