TDA4VM: First Hands on coding Edge AI Academy issue on TDA4VM-EVM

Hello Xavier,

Thank you for your interest and for trying out these examples.

I did a google translate and looks like "Kernel reboot" message.

DId you go through all the steps and compile the artifacts in step 3? Are you using those same artifacts in step 4?

Best regards,

Srik 

Hello Srik,

I apologize, I forgot to translate the error message on my screen.

It is a "kernel reboot" message saying : "Kernel Restarting.    The kernel appears to have died. It will restart automatically."

Yes I have done all the steps and I am using the same atifacts in step 4 that I have compiled in step 3. 

Does this error can occur because I am using this SDK : PROCESSOR-SDK-LINUX-SK-TDA4VM_08.01.00.02 | TI.com ?

Can I use MMA with this SDK or do I need to build RTOS  SDK to use MMA? 

Best regards, 

Xavier

No problem. I was able to translate and figure this out,.

The inference step is working in Step 3 on the cloud tool, right? Step3 code does both compilation and inference on the Cloud tool.

The issue might be the mismatch between the Cloud tool version and the SDK version on the actual hardware as we recently had an SDK update on the SK EVM.

I'll check on this.  Thanks Srik

Hello Srik,

Yes on the cloud tool everything works perfectly but I would like to use my EVM.

I have been able to go through the first error, a misspelling on a folder name was making the kernel died. There is no longer error reported but a weird behaviour from the MMA accelerator please see attach picture:

This is obtain thanks to the 3_EdgeAI_Model_Compile_Run from /cfs-file/__key/communityserver-discussions-components-files/791/HelloWorld_5F00_PC_5F00_Cloud_5F00_TDA4VM_5F00_SK_5F00_EVM_5F00_v2.zip 

After compiling on the cloud I transfered the custom-artifacts folder to my EVM board and I run 4_Infer_SK_EVM_ARM_ONLY.ipynb.

Best regards, 

Xavier

Hi Xavier, This issue you are seeing on the SK EVM, right? On the Cloud, everything runs fine as you stared earlier. Is that correct? That might be because of the versions. I'll check. Are you able to run the out-of-the-box demos on the SK EVM? Thank you, Srik

Hi Srik,

Yes I am seeing this issue on the SK EVM. 

I succeeded this week to compile my "custom-artifacts" on my computer thanks to "edgeai-benchmark" and from these artifacts I have great result in the inferences on the EVM, however the artifacts compiled from the Cloud do not work as reported before.

If I compile and run everything on the Cloud it works but when the compilation is done on the Cloud and the inference on the EVM, MMA have the same behaviour reported on my previous message.

Yes I am OK with the out-of-the-box demos :) 

Thank you,

Best regards,

Xavier