https://github.com/scottlamb/moonfire-tflite
Rust wrapper around the TensorFlow Lite C API and edgetpu C API
https://github.com/scottlamb/moonfire-tflite
Last synced: over 1 year ago
JSON representation
Rust wrapper around the TensorFlow Lite C API and edgetpu C API
- Host: GitHub
- URL: https://github.com/scottlamb/moonfire-tflite
- Owner: scottlamb
- License: apache-2.0
- Created: 2020-03-26T14:21:16.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2023-01-09T11:19:12.000Z (over 3 years ago)
- Last Synced: 2025-04-05T01:51:12.505Z (over 1 year ago)
- Language: Rust
- Size: 6.83 MB
- Stars: 9
- Watchers: 3
- Forks: 3
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE.txt
Awesome Lists containing this project
README
# moonfire-tflite
This is a simple Rust wrapper around the [TensorFlow
Lite](https://www.tensorflow.org/lite) and
[edgetpu](https://github.com/google-coral/edgetpu) libraries, written by
Scott Lamb <slamb@slamb.org>. It's primarily made to support video
analytics in [Moonfire NVR](https://github.com/scottlamb/moonfire-nvr).
As compared to the [tflite crate](https://crates.io/crates/tflite), advantages:
* Because it wraps the [C
API](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/c)
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](https://github.com/google-coral/edgetpu/issues/44#issuecomment-589170013)
needed to work with the latest `edgetpu` library. (Adjusting to new
TensorFlow version is much easier because of the simpler API.)
* It wraps the `edgetpu` library 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.
## License
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.