net2i 2.4

NeT2i (Network to Image) converts network traffic data into RGB images suitable for CNN-based anomaly detection.

net2i - Network Data to Image Converter

A Python library for converting network traffic data (CSV format) into RGB images for machine learning applications, particularly CNNs. net2i uses lossless encoding to preserve all data information while creating CNN-ready image datasets from network traffic logs.

🔄 Companion Tool: Use i2net to decode images back to CSV format

🚀 Features

• 🔍 Automatic IP Version Detection: Separates IPv4 and IPv6 data automatically
• 💎 Lossless Data Encoding: Converts network data to RGB pixels without information loss
• 🌐 Multiple Data Type Support: Handles IP addresses, MAC addresses, floats, integers, and strings
• 🧠 CNN-Ready Output: Generates images optimized for convolutional neural networks
• 📋 Type Information Preservation: Saves encoding metadata for data reconstruction via i2net
• ⚙️ Configurable Parameters: Customizable image size and output directories
• 🔀 Mixed IP Version Support: Processes IPv4 and IPv6 data in the same dataset

📦 Installation

pip install pandas numpy pillow

Requirements:

• Python 3.9+
• pandas
• numpy
• Pillow (PIL)
• ipaddress (built-in)

🚀 Quick Start

Basic Usage

import net2i

# Convert network traffic CSV to images
results = net2i.encode('network_traffic.csv')
print(f"Generated {results['total_images']} images in '{results['output_dir']}'")

With Custom Configuration

import net2i

# Configure for specific CNN requirements
results = net2i.encode(
    'firewall_logs.csv',
    output_dir='cnn_training_data',
    image_size=224  # ResNet/VGG input size
)

Global Configuration

import net2i

# Set global defaults
net2i.set_config(
    output_dir='training_images',
    image_size=150,
    clean_existing=True
)

# Use configured settings
results = net2i.encode('network_data.csv')

📊 Supported Network Data Types

Data Type	Detection Method	Encoding Strategy	Output Pixels
IPv4 Address	Automatic pattern matching	Split into 4 octets → IEEE 754 encoding	8 RGB pixels
IPv6 Address	Automatic pattern matching	128-bit → 16 bytes + 2 padding	6 RGB pixels
MAC Address	Regex: XX:XX:XX:XX:XX:XX	Split into 2 hex chunks → float encoding	4 RGB pixels
Float/Integer	Numeric detection	Direct IEEE 754 encoding	2 RGB pixels
String	Default fallback	Consistent hash → float encoding	2 RGB pixels

Encoding Details

• Two-Pixel-Per-Float Strategy: Each float value uses exactly 2 RGB pixels (6 bytes) for lossless IEEE 754 representation
• IP Address Decomposition: IPv4 addresses split into octets, IPv6 addresses use full 128-bit representation
• Hash-Based String Encoding: Strings converted using consistent hashing for reproducible results

🔧 API Reference

Core Functions

encode(csv_path, **kwargs)

Main function to convert CSV network data to images.

Parameters:

• csv_path (str): Path to input CSV file containing network traffic data
• output_dir (str, optional): Directory for output images (default: 'data')
• image_size (int, optional): Size of square output images (default: 150)

Returns:

{
    'input_file': 'network_traffic.csv',
    'output_dir': 'data',
    'image_size': 150,
    'has_ipv4': True,
    'has_ipv6': False,
    'total_images': 1000,
    'ipv4_results': {...},
    'ipv6_results': None
}

load_csv(csv_path)

Load and validate network traffic CSV file.

set_config(**kwargs)

Configure global settings for all operations.

Configuration Options:

• output_dir: Output directory for generated images
• image_size: Image dimensions (width × height) - tune for your CNN architecture
• types_file: JSON file for IPv4 type information ('data_types.json')
• types_file_ipv6: JSON file for IPv6 type information ('data_types_ipv6.json')
• clean_existing: Clean existing files before processing (default: True)

Utility Functions

• show_config(): Display current configuration
• reset_config(): Reset to default settings
• help(): Show detailed usage examples

📁 Output Structure

Generated Files

output_dir/
├── ipv4_0.png              
├── ipv4_1.png
├── ipv4_2.png
├── ...
├── ipv6_0.png              
├── ipv6_1.png
├── ...
data_types.json             
data_types_ipv6.json        
ipv4_rows.csv              
ipv6_rows.csv              

Type Information Files

{
  "ip_version": "IPv4",
  "original_types": ["IPv4 Address", "Float", "String"],
  "final_types": ["IPv4 Address", "IPv4 Address", "IPv4 Address", "IPv4 Address", "Float", "String"],
  "encoding_info": {
    "description": "Data type mapping for decoding - IPv4 version",
    "float_encoding": "Each float becomes 2 RGB pixels (6 bytes total)",
    "ipv4_encoding": "IPv4 address split into 4 octets, each becomes 2 RGB pixels"
  },
  "original_columns": 3,
  "final_columns": 6
}

🧠 Machine Learning Integration

TensorFlow/Keras Pipeline

import net2i
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator

net2i.set_config(image_size=224, output_dir='training_data')
results = net2i.encode('network_traffic.csv')

datagen = ImageDataGenerator(
    rescale=1./255,
    validation_split=0.2,
    rotation_range=10,
    width_shift_range=0.1,
    height_shift_range=0.1
)

train_generator = datagen.flow_from_directory(
    'training_data',
    target_size=(224, 224),
    batch_size=32,
    class_mode='categorical',
    subset='training'
)

validation_generator = datagen.flow_from_directory(
    'training_data',
    target_size=(224, 224),
    batch_size=32,
    class_mode='categorical',
    subset='validation'
)

PyTorch Integration

import net2i
import torch
from torchvision import transforms, datasets
from torch.utils.data import DataLoader

net2i.encode('network_logs.csv', image_size=224)

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

dataset = datasets.ImageFolder('data', transform=transform)
dataloader = DataLoader(dataset, batch_size=32, shuffle=True, num_workers=4)

📋 Input Data Format

CSV Structure

• No headers required
• Mixed data types supported
• Standard network formats: IPs, MACs, ports, timestamps

Example CSV

12,2001:0db8:85a3:0000:0000:8a2e:0370:7334,52:54:00:34:65:b2,...
11,192.168.248.159,52:54:00:34:65:b2,...
12,192.168.248.159,52:54:00:34:65:b2,...

Sources

• Firewall logs
• IDS/IPS alerts
• Flow records
• Packet capture summaries
• 5G-MEC data
• IoT communications

🔄 Integration with i2net

import net2i
results = net2i.encode('original_traffic.csv')

import i2net.decoder as decoder
decoded_results = decoder.load_data('data', 'reconstructed_traffic.csv')

🛠️ Technical Implementation

1. Load CSV and detect column types
2. Separate IPv4 and IPv6 rows
3. Encode each value by type (MAC/IP/float/string)
4. Use IEEE 754 encoding for floats
5. Map values to RGB pixels
6. Assemble square images

🎯 Image Size Recommendations

CNN Architecture	Recommended Size
Default	150×150
ResNet, VGG	224×224
Inception	299×299
Custom	Match your model

🚨 Troubleshooting

• No IPs detected: Check your CSV contains valid IPs
• Corrupt images: Set clean_existing=True
• Memory errors: Lower image size or batch processing

🖥️ CLI Usage

python net2i.py network_traffic.csv
python net2i.py firewall_logs.csv cnn_images 224
python net2i.py  # Show help

📚 Citation

@inproceedings{fernando2023new,
  title={New algorithms for the detection of malicious traffic in 5g-mec},
  author={Fernando, Omesh A and Xiao, Hannan and Spring, Joseph},
  booktitle={2023 IEEE Wireless Communications and Networking Conference (WCNC)},
  pages={1--6},
  year={2023},
  organization={IEEE}
}

👥 Author

• Omesh Fernando

📄 License

MIT License

🔗 Related

• i2net: Reverse net2i images
• IEEE WCNC 2023 Paper

🤝 Contributing

1. Fork repo
2. Create a branch
3. Add tests
4. Commit
5. Push
6. PR

💬 Support

• GitHub Issues
• net2i.help()
• i2net

---

🔄 Tip: net2i images are lossless and fully reversible with i2net.