SDK Development flow¶
Processor SDK Linux for Edge AI comprises of 3 parts,
Edge AI Applications
Processor SDK - Linux
Processor SDK - RTOS
If you wish to work with just the Edge AI Applications, you can edit-build-run directly on the target by flashing the etcher image as specified in Preparing SD card image. No need to build the RTOS binaries as described below!
If you are modifying any RTOS components such as vision_apps, tiovx, tidl, tiadalg etc. you will need to rebuilt the RTOS binaries and update the SD card filesystem with the latest. This requires the Processor SDK RTOS source package and Processor SDK Linux filesystem packages.
If you are modifying any Linux DTBs, updating u-boot, updating system firmware (sysfw), device driver and so on, you will need the Processor SDK Linux source package.
Getting PSDK RTOS¶
You can download the latest Processor SDK RTOS for TDA4VM from the below links,
The Processor SDK RTOS source package:
The Processor SDK Linux filesystem and BOOT partition files are also required:
Building the RTOS SDK from source¶
Untar the ti-processor-sdk-rtos-j721e-evm-08_05_00_xx.tar.gz to some location
on PC and copy the tisdk-default-image-j7-evm.tar.xz and boot-j7-evm.tar.gz
into the RTOS SDK folder. Execute the
setup_psdk_rtos.sh script from the
folder below. This will untar the linux filesystem, download GCC cross compile
toolchain and dependencies all in preparation to build the SDK
#Run the setup script as shown <PSDK-RTOS path>#./psdk_rtos/scripts/setup_psdk_rtos.sh
Ensure that you are building only for the
target and the profile is set to
release mode. In the file
# Build for SoC BUILD_TARGET_MODE?=yes # Build for x86 PC BUILD_EMULATION_MODE?=no # valid values: release debug all PROFILE?=release
Navigate to vision_apps folder and build the RTOS components
<PSDK-RTOS path>/vision_apps#BUILD_EDGEAI=yes make sdk -j8
Once the build is complete with no errors install the Linux filesystem to an SD card. A minimum of 16GB is highly recommended. Insert an SD card and unmount its partitions. If you had previously installed using the same SD card, you might have to unmount the BOOT and rootfs partitions
/dev/sda2 are just examples here, and the SD card might
be mounted differently on your system. Ensure that you are unmounting the
correct partitions before reformatting the SD card!
umount /dev/sda1 umount /dev/sda2
Use the script to create required partitions and filesystem format. This will create two partitions namely BOOT and rootfs
<PSDK-RTOS path>#sudo ./psdk_rtos/scripts/mk-linux-card.sh /dev/sda
Follow the interactive menu (mostly press Y and Y) to correctly format the SD card for Linux boot. Once successfully complete, PLEASE EJECT THE SD CARD AND INSERT AGAIN. This ensures that files are not installed under root permissions and the following steps do not require sudo permissions
Use another script to install the filesystem to the rootfs partition
Next transfer the RTOS component libraries, header files and remote core binaries to the SD card.
<PSDK-RTOS path>/vision_apps# BUILD_EDGEAI=yes make linux_fs_install_sd
Copies the vision_apps headers under
Copies the RTOS binaries under
Getting PSDK LINUX¶
You can skip this step if you don’t intend to change anything on the Linux kernel, u-boot or file-system
You can download the latest Processor SDK Linux for TDA4VM from the below link,
Please refer to Processor SDK Linux documentation for installation steps.
Building the LINUX SDK from source¶
Please refer to Simplified SDK Build for quick changes and build in u-boot, sysfw (system firmware) and linux kernel.
Please refer to Building SDK with Yocto for building each individual components. This will require a longer time to build and a powerful machine.
** This completes the PSDK Linux and RTOS setup **
At this point the SD card is ready with Linux image and RTOS components.
Next we need to clone the target side components under SD card
If the provided steps are not executing you might have to provide permissions
for the scripts to make directory under SD card /opt.
You need to change permissions on SD card /opt folder NOT your Linux PC
/opt folder. Change permissions of SD card /opt directory
/media/<user-name>/rootfs# sudo chmod -R ugo+w opt
First clone the
edge_ai_apps repo into the SD card under
/media/<user-name>/rootfs/opt#git clone --single-branch --branch master git://git.ti.com/edgeai/edge_ai_apps.git
Once edge_ai_apps is cloned, on the PC itself run the
This will clone and install DL dependencies, clone the edgeai-gst-plugins and
edgeai-tiovx-modules repo on the SD card /opt.
Also download the recommended set of models or any model that you would like
to test by running the
** This completes the setup required on PC side **
Before ejecting the SD card issue a ‘sync’ on the terminal to commit all changes.
Target side steps¶
Boot the SK with the prepared SD card. Make sure you have connected the
camera and display. After boot, login as ‘root’ with no password. If you have
flashed the SD card using Balena etcher tool, then upon login you will notice
the wallpaper displayed on the screen disappears and comes back. The prompt will
also change to
/opt/edge_ai_apps#. This is done by automatically calling the
If you are building from scratch, upon login you will notice that the wallpaper is still displayed on the screen. The init_script.sh is not automatically called during login. You will have to manually navigate to edge_ai_apps repo and execute every time it boots or call it under .profile
Only for first time, you will have to run the setup_script.sh once again. This time it will not download the dependencies but compile and install the edgeai-gst-plugins, edgeai-tiovx-modules, edgeai-dl-inferer and apps_cpp demos.
For more information about what init_script does, visit init_script
Now you are ready to run the Python and C++ demos!
IMPORTANT: Due to differences in GLIB version between Yocto and Docker you will have to rebuild the GST plugins and C++ apps when you make the switch. So always call the setup_script.sh to rebuild the plugins and C++ apps for Yocto/Docker while switching.