深度学习模型加密解密并附图

## TFSecured Small library for tensorflow proto models (*.pb) encryption/decryption. ## AES You may use random string with random length like a key, then library calculates sha256 hash of it and uses as internal key with size 256 bits. ## Usage Copy sources from [TFSecured](https://github.com/dneprDroid/TFSecured/tree/master/TFSecured) dir into your project. C++ Usage (see [TFPredictor.mm](https://github.com/dneprDroid/TFSecured/blob/master/iosDemo/TFSecured-iOS/tf/predict/TFPredictor.mm)): ```cpp #include <GraphDefDecryptor.hpp> ........ tensorflow::GraphDef graph; // Decryption: const std::string key = "JHEW8F7FE6F8E76W8F687WE6F8W8EF5"; auto status = tfsecured::GraphDefDecryptAES(path, // path to *.pb file (frozen graph) &graph, key); // your key if (!status.ok()) { std::cout << status.error_message() << std::endl; return; } // Create session : std::unique_ptr<Session> session(NewSession(options)); status = session->Create(graph); // Run session .... ``` Encrypt model via python script (see [encrypt_model.py](https://github.com/dneprDroid/TFSecured/blob/master/python/encrypt_model.py)): ```bash $ python encrypt_model.py <INPUT_PB_MODEL> \ <OUTPUT_PB_MODEL> \ <KEY> # optional, generated randomly by script ``` ## iOS Demo (Digit recognizer) Prepare repository (install Tensorflow via pods) ```bash $ cd iosDemo $ pod install ``` And open it in Xcode.