README.md

The goal of this project is to provide neural network architectures for traffic classification and their pre-trained weights.

The package provides two network architectures, 30pktTCNET and Multi-modal CESNET v2, both visualized in the following pictures. See the getting started page and models reference for more information.

🐸 🐸 See a related project CESNET DataZoo providing large TLS and QUIC traffic datasets. 🐸 🐸

📓 📓 Example Jupyter notebooks are included in a separate Traffic Classification Examples repository. 📓 📓

🚀 🚀 Transfer Learning Codebase for reproducing experiments from our paper — covering ten downstream traffic classification tasks with three transfer approaches (k-NN, linear probing, and full model fine-tuning). 🚀 🚀

30pktTCNET

Multi-modal CESNET v2

Installation

Install the package from pip with:

pip install cesnet-models

or for editable install with:

pip install -e git+https://github.com/CESNET/cesnet-models

Papers

Models from the following papers are included:

• Universal Embedding Function for Traffic Classification via QUIC Domain Recognition Pretraining: A Transfer Learning Success
  Jan Luxemburk, Karel Hynek, Richard Plný, and Tomáš Čejka
  IEEE Transactions on Network and Service Management, 2025

• Encrypted traffic classification: the QUIC case
  Jan Luxemburk and Karel Hynek
  2023 7th Network Traffic Measurement and Analysis Conference (TMA)

• Fine-grained TLS services classification with reject option
  Jan Luxemburk and Tomáš Čejka
  Computer Networks, 2023