Skip to main content

Image Recognition

This example demonstrates how you can deploy a function that performs secure image recognition using a TensorFlow Lite model. By using Cape, the model is protected because it can only be accessed within a secure enclave. The input data is also protected because it is encrypted prior to being uploaded, and is sent over TLS into the Cape enclave.

Code

Clone the capeprivacy/functions public repository:

git clone git@github.com:capeprivacy/functions.git

Navigate into the folder containing the code for the image recognition function:

cd image_recognition

Install the required dependencies into a target folder (named deploy in our example):

docker run -v `pwd`:/build -w /build --rm -it python:3.9-slim-bullseye pip install -r requirements.txt --target ./deploy/

Note: sudo may be required when running this command.

Login

Log in to Cape using:

cape login

Which produces:

Your CLI confirmation code is: <SOME_CODE>
Visit this URL to complete the login process: https://login.capeprivacy.com/activate?user_code=<SOME_CODE>
Congratulations, you're all set!

Deploy

Deploy your function to Cape using:

cape deploy ./deploy

Which produces:

Deploying function to Cape ...
Success! Deployed function to Cape
Function ID ➜ jjAxk3CFnRx7whdi5oPffQ
Checksum ➜ 00f617b57eebfaa29b7f091b3dab104fc690b2173d0de8fa46dc5d426bf6a2e1

Run

Invoke your function using:

cape run jjAxk3CFnRx7whdi5oPffQ -f cat.jpg

Which produces:

Image Label is : 'Egyptian cat' with Accuracy : 72.27 %.

Feel free to try with other images:

cape run jjAxk3CFnRx7whdi5oPffQ -f coffee.jpg
cape run jjAxk3CFnRx7whdi5oPffQ -f lizzard.jpg