I would like to know how to generate a TFLite model that can run properly on a specific platform (e.g., Genio 510/Genio 1200..).
Are there any recommended workflows or best practices to ensure platform compatibility during the conversion process from a trained TensorFlow model to TFLite?
Preferred method to convert Tensorflow Model to TFLite is by using tensorflowlite converter.
Once you have the Tensorflow saved_model, you can convert it to TFLite using the method mentioned here
We do need to optimize these models to work best on target. For this, we quantize the models using either Quantize Aware Training (QAT), or Post Training Quantization (PTQ).
More information on this can be found here
We also do provide mtk_converter tools to convert the model to TFLite, but this converter might add mtk custom ops at times and then you cannot use online pathway for model execution. You’d have to execute such TFLite models via offline pathway only, by converting it to DLA and then using neuronrt for execution