metadata
license: mit
datasets:
- flwrlabs/celeba
metrics:
- accuracy
base_model:
- microsoft/resnet-18
Breaking DataMix – Pretrained Models
This repository provides pretrained models used in our study on inversion attacks against DataMix.
Overview
The models are based on a ResNet-18 architecture and are trained on a mixed version of the CelebA dataset. They are specifically designed for evaluating privacy vulnerabilities and recovering private mixing coefficients via inversion attacks.
For full details about the methodology, training setup, and data preprocessing, please refer to the main project repository:
👉 https://github.com/hehteram/Breaking-DataMix
Contents
- Pretrained model checkpoints (.pth)
- Variants corresponding to different experimental settings
Usage
from huggingface_hub import hf_hub_download
import torch
path = hf_hub_download(
repo_id="hehteram/Breaking-DataMix",
filename="models/model.pth"
)
model.load_state_dict(torch.load(path))
Notes
- These models are provided for research and reproducibility purposes only.
- The dataset used is derived from CelebA with specific modifications (see project repository for details).
License
Please refer to the main GitHub repository for licensing and usage terms.