TensorFlow Model Analyzer

A command-line tool for analyzing TensorFlow models and extracting basic information about their architecture.

Features

• Analyze TensorFlow model files (.h5, .pb, SavedModel)
• Extract key information:
  • Number of layers
  • Layer types
  • Total parameter count
  • Input/output shapes
• Export analysis to TXT or JSON format
• Simple command-line interface

Installation

From Source

git clone https://github.com/lognjen/tensorflow-model-analyzer.git
cd tensorflow-model-analyzer
pip install -e .

Requirements

• Python 3.7+
• TensorFlow 2.x

Usage

Basic Usage

# Analyze a model and print to console
tf-analyzer path/to/model.h5

# Analyze a SavedModel directory
tf-analyzer path/to/saved_model_dir

# Save output to TXT file
tf-analyzer path/to/model.h5 --output model_info.txt

# Save output to JSON file
tf-analyzer path/to/model.h5 --output model_info.json --format json

Options

--output, -o     Output file path (default: print to console)
--format, -f     Output format: 'txt' or 'json' (default: determined by file extension)
--verbose, -v    Include additional model details
--help, -h       Show help message

Examples

Example Output (TXT format)

TensorFlow Model Analysis
========================
Model: my_model.h5
Date: 2025-04-10 14:30:22

Summary:
- Total layers: 15
- Trainable parameters: 1,435,788
- Non-trainable parameters: 256

Layer Types:
- Conv2D: 8
- BatchNormalization: 3
- MaxPooling2D: 2
- Dense: 2

Input Shape: (None, 224, 224, 3)
Output Shape: (None, 10)

Example Output (JSON format)

{
  "model_info": {
    "filename": "my_model.h5",
    "analysis_date": "2025-04-10 14:30:22"
  },
  "summary": {
    "total_layers": 15,
    "trainable_parameters": 1435788,
    "non_trainable_parameters": 256
  },
  "layer_types": {
    "Conv2D": 8,
    "BatchNormalization": 3,
    "MaxPooling2D": 2,
    "Dense": 2
  },
  "shapes": {
    "input": "(None, 224, 224, 3)",
    "output": "(None, 10)"
  }
}

License

This project is licensed under the MIT License - see the LICENSE file for details.