The first step of using TFLite is choosing a model to convert and use. This includes using pre-trained models, custom-trained models, or fine-tuned models. The TFLite team provides a set of pre-trained and pre-converted models that solve a variety of machine learning problems. These include image classification, object detection, smart reply, pose estimation, and segmentation. Using fine-tuned models or custom-trained models requires another step where they are converted into TFLite format.
TFLite is designed to execute models efficiently on devices, and some of this efficiency comes inherently from the special format used to store the models. TF models must be converted into this format before they can be used in TFLite. Converting models reduces file size and add optimizations that don't affect accuracy. Other, more lossy, optimizations can be...
