This is a simple Rust wrapper around the TensorFlow Lite and edgetpu libraries, written by Scott Lamb <slamb@slamb.org>. It's primarily made to support video analytics in Moonfire NVR.
As compared to the tflite crate, advantages:
- Because it wraps the C API rather than the C++ API, it's simpler and quicker to build. It doesn't need bindgen. (This is the primary reason I wrote my own.)
- It runs with a more modern version of TensorFlow, including the specific
commit
needed to work with the latest
edgetpulibrary. (Adjusting to new TensorFlow version is much easier because of the simpler API.) - It wraps the
edgetpulibrary as well.
Disadvantages:
- It's much less mature: less usage, no documentation, no CI.
- It's less feature-rich; the C API can't do everything the C++ API can.
Apache-2.0. I'd like to dual-license with MIT, but the stock models I'm using for test data are Apache-licensed. Eventually I will find or make new test data models and relicense.