FlowRefiner

This repository contains the code for the paper: FlowRefiner: A Robust Traffic Classification Framework against Label Noise (NeurIPS 2025).

Setup

Dependencies:

• Python ≥ 3.8
• PyTorch ≥ 1.10
• torchvision
• timm==0.3.2
• scikit-learn
• matplotlib

Dataset

The dataset format is consistent with the YaTC:

[Dataset_Path]/[train/test]/[Class]/[Sample]

Noisy dataset example (ISCXVPN with 20% label noise): Link

Path: ./data/ISCXVPN_noisy20

To generate a dataset with a specified label noise ratio:

Code: data_process.py
Function: make_noise_dataset(raw_dataset_path, noise_ratio)

Run

Pre-trained encoder: Link

Path: ./output_dir/pretrained_encoder.pth

Run FlowRefiner:

python FlowRefiner_train.py 

Contact-Info

Link to our laboratory: SJTU-NSSL

Mingwei Zhan Email: mw.zhan@sjtu.edu.cn

Ruijie Zhao Email: ruijiezhao@seu.edu.cn