DEEPCRAFT™ Model Converter

DEEPCRAFT™ Model Converter is a comprehensive solution designed to facilitate the conversion of pre-trained models for deployment on an Infineon target platforms. By leveraging Model Converter, users can generate code for existing models, optimize pre-trained models for specific target devices through configurable optimization parameters, and validate model performance on the desktop prior to actual device deployment.

DEEPCRAFT™ Model Converter is available in both Graphical User Interface (GUI) and Command-Line Interface (CLI) versions, providing flexibility and convenience for developers working in a variety of workflows. Model Converter is fully compatible with Windows, macOS, and Linux operating systems allowing users to integrate the models into the preferred development environments. Whether you are a researcher or a developer, DEEPCRAFT™ Model Converter simplifies the process of importing, optimizing, and validating your pre-trained models for Infineon devices.

Key Features

Understanding DEEPCRAFT™ Model Converter

The intended use of DEEPCRAFT™ Model Converter is to generate code files that enable the execution of Neural Networks (in .h5, .tflite, .pt2 formats) on an an Infineon PSOC™ 6 or PSOC™ Edge target. The Model Converter accepts the network file, along with optional calibration or validation data, as input and generates model.c and model.h files that can be imported into a ModusToolbox™ project to perform inference (with optimizations for the selected target SoC). This workflow assumes that you already have implemented the necessary pre and post-processing steps, as well as any required sensor data acquisition code, to ensure that the data is in the proper format for passing to the network. The generated code depends on the following ModusToolbox™ assets or libraries that can be imported through the ModusToolbox™ Eclipse IDE library manager.For Infineon PSOC™ 6 target, the generated code relies on ml-middleware 3.0.1 and ml-tflite-micro 3.0.1 libraries, while for PSOC™ Edge target, the generated code relies on ml-middleware 3.1.0 and ml-tflite-micro 3.1.0 libraries,understanding these dependencies is crucial for successful model deployment and execution.

Supported Models

You can generate code for the TensorFlow, Keras or PyTorch models with the following specifications:

TensorFlow or Keras

You can use TensorFlow models built with Keras 2 saved in (.h5) format. The models exported with other Keras 3 are not supported. If you use TensorFlow 2.16+ to build a model: install the tf-keras python package, force Keras 2 with os.environ["TF_USE_LEGACY_KERAS"] = "1" before the statement importing Tensorflow or Keras into your script.

import os
os.environ["TF_USE_LEGACY_KERAS"] = "1"
import tensorflow as tf
from tensorflow import keras

PyTorch

You must use torch==2.6.0, when exporting models to the .pt2 format. Models exported with other PyTorch versions are not supported. For information on exporting pytorch models to .pt2, refer to torch.export (opens in a new tab) and serialization (opens in a new tab) section in the PyTorch documentation.

Whats Next?

Let's install the DEEPCRAFT™ Model Converter and launch the Graphical User Interface and Command Line Interface. Refer to Install and Launch DEEPCRAFT™ Model Converter.