mp-tensorflow is a package for secure tensorflow inference that utilizes the MP-SPDZ framework. For all supported models, computation are preformed in four parts:
- Pretrained model is downloaded
- Resource/Dependency collection
- Data processing
- MP-SPDZ compilation
Supported models:
Before running Tensorflow inference, be sure to run the compilation instructions once:
Scripts/setup-ssl.sh
make -j 8 tldr
make emulate.x
Python version must be between 3.5 - 3.7.
Additional dependencies may be installed through requirements.txt:
cd TensorflowInf
pip install -r requirements.txt
WARNING: installing through requirements.txt may downgrade some Python libraries. For that reason, it is recommended to use virtualenv to isolate pip installs.
cd TensorflowInf/SqueezeNet
./tf-inference.sh -i SampleImages/sample_image_1.JPEG -n 1
-i <INPUT_IMAGE_FILE> Path to image file to run computation on. This may be relative or absolute path.
-n <NUMBER_OF_THREADS> Number of threads for compilation
cd TensorflowInf/ResNet
./tf-inference.sh -i SampleImages/sample_image_1.JPEG -n 1
-i <INPUT_IMAGE_FILE> Path to image file to run computation on. This may be relative or absolute path.
-n <NUMBER_OF_THREADS> Number of threads for compilation