yechanp / Membership-Inference-Attacks-Against-Object-Detection-Models

Membership-Inference-Attacks-Against-Object-Detection-Models

This is the codes for https://arxiv.org/abs/2001.04011 (Membership Inference Attacks Against Object Detection Models).

Requirements

• Python
• Chainer
• Pytorch

Training Shadow and Target Models

To train object detection models, please run the following files.

# For SSD300 using VOC dataset
python train_Chainer_ssd_voc_shadow.py  [gpu_id]
python train_Chainer_ssd_voc_target.py  [gpu_id]

To train the attacker using the shadow model with the shadow dataset, please execute

python mia_train_attacker_.py  config.py

To attack the target object detection model, please execute

python mia_evaluate.py  config.py

Trained Object Detection Model Weights

For VOC-SSD300 models, you can download the weights of the shadow and target models at https://drive.google.com/drive/folders/1E45q9K1kqLzVqyYrrpZid07oq4JN5Z-V?usp=sharing .

Futher model weights will be uploaded later.