MATLAB Coder Support Package for PyTorch® and LiteRT Models enables you to generate readable and portable C/C++ source code from a variety of pretrained PyTorch and LiteRT (formerly TensorFlow Lite) deep learning networks. You can integrate the complete application, including MATLAB and Simulink pre and postprocessing and deep learning models, into your existing C/C++ projects as source code, static libraries, or dynamic libraries. You can then deploy the complete application onto various hardware platforms, from desktop systems to embedded hardware.
The support package also enables you to generate CUDA® code from PyTorch and LiteRT models for modern NVIDIA® GPUs, including those embedded on NVIDIA Jetson™ and NVIDIA Clara™ platforms (with GPU Coder). You can apply code customizations, hardware-specific optimizations, and code verification using software-in-the-loop (SIL) and processor-in-the-loop (PIL) testing (with Embedded Coder). You can also generate code from Simulink models that contain the PyTorch and LiteRT models (with Simulink Coder).
Product Highlights
Generate Source Code
Generate readable and portable C/C++ source code from pretrained PyTorch and LiteRT models, including the preprocessing and postprocessing logic. Deploy the royalty-free generated code in commercial applications to your customers at no charge. Alternatively, use processor-specific intrinsics for select targets, including Infineon© AURIX™ TC4x microcontrollers (with Embedded Coder).
Run on Hardware
Compile the generated code and execute it on any hardware supported by your compiler, from desktop systems to embedded processors and microcontrollers. Target any device by manually integrating the generated code with your application. Automate the process for select targets using hardware support packages, including MATLAB Support Package for Raspberry Pi and MATLAB Coder Support Package for NVIDIA Jetson and NVIDIA DRIVE Platforms.
Optimize Generated Code
Generate single instruction, multiple data (SIMD) instructions and parallelize loops with multicore OpenMP code. Generate processor-specific intrinsic code with Embedded Coder for select targets, including ARM® Cortex®-M, Cortex-A, and Infineon AURIX TC4x microcontrollers for faster execution.
Deep Learning Applications
Use deep learning networks from PyTorch and LiteRT to process different types of data, and deploy the networks to resource-constrained hardware.
Signal and Time-Series Data
Use deep learning on signal and time-series data for tasks such as anomaly detection, signal classification, battery state-of-charge estimation, and speech recognition. Generate code and deploy networks based on recurrent neural networks (RNNs), long short-term memory (LSTM)/gated recurrent units (GRUs), transformers, and autoencoders to CPUs (including ARM Cortex-M, Cortex-A, and Intel®) and NVIDIA GPUs.
Image and Video Data
Use deep learning on images and videos to classify images, detect and track objects within the frame, and segment image and frame regions into different types of objects. Generate code and deploy networks such as YOLO11, DINOv2, and Depth-Anything to CPUs (including ARM Cortex-A and Intel) and NVIDIA GPUs.
Request Support Package
Talk to a technical expert.
