зеркало из https://github.com/microsoft/EdgeML.git
061881b854
* Adding keras inference layer to be used along with tflite API proposed by Tensorflow in the 2.0 version. This helps to convert the pre-trained ProtoNN and Bonsai models into the tflite format and shows how to use this model for prediction using the tflite interpreter. Note that the model training still happens on the Tensorflow-1.15 version (this can be changed later in which case you'll be left with only the Bonsai and ProtoNN Keras layers). For using these utilities you will need to use the tensorflow-2.X for conversion. * Added instructions to use the tflite model converter utilities and the Keras layers. Also updated the convert utility to include the quantization as an optimization step enabled by default. This also includes a bug fix in the requirements-gpu file which incorrectly uses an older version of the numpy library incompatible with the tensorflow version 1.15 * Included assertion that the converter utility works only on Tensorflow 2.X API. |
||
|---|---|---|
| .. | ||
| Bonsai | ||
| EMI-RNN | ||
| FastCells | ||
| ProtoNN | ||