classicvalues / edgeai-tidl-tools

Edgeai TIDL Tools and Examples - This repository contains Tools and example developed for Deep learning runtime (DLRT) offering provided by TI’s edge AI solutions.

EdgeAI TIDL Tools and Examples

This repository contains examples developed for Deep learning runtime (DLRT) offering provided by TI’s EdgeAI solutions. This repository also contains tools that can help in deploying AI applications on TI’s EdgeAI solutions quickly to achieve most optimal performance.

• EdgeAI TIDL Tools and Examples
  • Introduction
  • Setup
    • Advanced Setup Options
  • Python Examples
  • CPP Examples
  • Jupyter Notebooks
  • Versioning
  • Notes
  • License

Introduction

The following sections describes the steps to install this repository, dependent components on your device and run the examples on the same. Most the steps explained in this page are common for PC emulation and execution on target. If any of the steps is different between PC and target, then same is called out in this document.

Setup

• This repository is validated on Ubuntu 18.04 in PC emulation mode and TDA4VM EVM using PSDK-RTOS release
• This repository works only with python 3.6 on PC (Which is default in Ubuntu 18.04)
• We have also validated under docker container in PC. Refer Dockerfile for the list of dependencies installed on top of ubuntu 18.04 base line
• Run the below script to install the dependent components on your machine and set all the required environments

git clone https://github.com/TexasInstruments/edgeai-tidl-tools.git
cd edgeai-tidl-tools
source ./setup.sh

Advanced Setup Options

• If you are planning to validate only python examples and avoid running CPP examples, invoke the setup script with below option

 source ./setup.sh --skip_cpp_deps

• If you have the ARM GCC tools chain already installed in your machine, invoke the setup script with below option and set "ARM64_GCC_PATH" environment variable

export ARM64_GCC_PATH=$(pwd)/gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu
source ./setup.sh --skip_arm_gcc_download

• If you are building the PSDK-RTOS from source and updating any of the TIDL tools during the development, then set "TIDL_TOOLS_PATH" environment variable before starting setup script

export TIDL_TOOLS_PATH=$PSDKR_INSTALL_PATH/tidl_j7_xx_xx_xx_xx/tidl_tools
source ./setup.sh

Python Examples

• Run below script to validate all the python examples available in the repository. This script would run both model-compilation and inference when executed on PC

./scripts/run_python_examples.sh

• This script would run only inference of example models when executed on target device like J7ES EVM or SK. So this script must be first executed on PC to generate the artifacts needed for inference and then copy below folders from PC to target device before running this script on device

./model-artifacts
./models

• Refer Python Examples for details on the custom model compilation and inference python examples

CPP Examples

• CPP APIs of the DL runtime offered by solutions only supports the model inference. So the user is expected to run the Python Examples on PC to generate the model artifacts.

• Refer CPP Examples for detailed instructions on building and running the CPP examples

Jupyter Notebooks

• All the noteboks can be executed in PC emulation mode, but only inference notebooks can be executed on target device.

• Run the below command to launch the Jupyter notebooks session

  cd examples/jupyter_notebooks
  source ./launch_notebook.sh
  

• Refer Jupyter Notebook for details on using Jupyter Notebooks examples

Versioning

• This repository would be tagged with same version as PSDK-RTOS for every release. For example 08.00.00.12.
• If there is any PC tools update/Bugfix which is compatible with a PSDK-RTOS version, then the same would be tagged with a additional version digits like 08.00.00.12.01
• Always refer the setup for current compatible PSDK-RTOS version

Notes

• These examples are only for basic functionally testing and performance benchmarking (latency and memory bandwidth). Accuracy of the models can be benchmarked using the python module released here edgeai-benchmark

License

Please see the license under which this repository is made available: LICENSE