The following provides a brief description of some of the more common concepts used by the MLTK.
.tflite: Tensorflow-Lite Model¶
.h5: Keras Model¶
This is generated by Keras after model training completes. This typically contains float32 weights. The TF-Lite converter uses this to generate a quantize
You may view the contents of this file by dragging and dropping into the webpage: https://netron.app.
.mltk.zip: MLTK Model Archive¶
This contains the model specification, trained model files (
.h5), training logs, and evaluate logs.
See Model Archive for more details.
.py: Model Specification¶
This defines the model structure, the training dataset including any augmentations, the training parameters, and any additional model parameters.
See Model Specification for more details.
Model Object Types¶
The following model Python objects are used by the MLTK:
The MltkModel contains all information required to train a model.
The TfliteModel loads a
.tflite model file and provides programmic access to it contents.
The KerasModel is what is trained by Tensorflow. This defines the actual model layout.