TDA4VM: How to display a model with two outputs on TDA4VM?

Hi,

Could you enable the remote side logs too?

Before running your application, please run "source ./vision_apps_init.sh" in order to enable the remote logs in your application.
Once this is done, please share your application logs as well as the logs of "source ./vision_apps_init.sh".

Regards,
Nikhil

Thanks for your answer.

I have some questions.

1. What are the remote side logs and application logs? Are they TDA4VM and PC?

2. My model has two outputs(object detection, segmentation). Object detection and segmentation have different post-processing methods. Is it necessary to modify vision_apps a lot to apply other post-processing to the output of the model?

3. Is there any way to debug in TDA4VM? As stated in the documentation, additional packages are required to build with PROFILE=debug. I applied for an additional package, but it was not approved. Can't debug mode be used without additional packages?

Thank you.

Hi,

Jaemin Lee said:

What are the remote side logs and application logs

Remote logs are the logs that are coming from the remote cores such as [MCU2_0], [MCU2_1] or in your case, may be some errors would have come from [C7x].
In order to view these logs, you would have to run source ./vision_apps_init.sh before running your application.

You have already provided a part of the application log in the screenshot attached in the description, but it would be helpful if I get the full logs along with the remote side logs enabled too on TDA4VM.

Jaemin Lee said:

Is it necessary to modify vision_apps a lot to apply other post-processing

May I know currently which demo are you running? From your description, your graph could look similar to this demo
Vision Apps User Guide: Auto Valet Parking Application 2

Or is it some other demo which you have modified?

Jaemin Lee said:

 Is there any way to debug in TDA4VM

You could use Code Composer studio (CCS) in order to debug on TDA4VM
6. CCS Setup for J721E — Processor SDK RTOS J721E (ti.com)

Regards,
Nikhil

Hi,

I have an additional question for the second answer.
Auto Valet Parking Application 2 is similar to what I was thinking. However, when inputting Auto Valet Parking Application 2, bin files of a completely different model are entered as input. For example:

However, the model I have is different from the example above. It is one model and has two outputs. My model has one input, and the output of my model has the results of object detection and segmentation, respectively.

The part of the code I wanted to modify is as follows.

In the picture above, I thought obj->tidlObj.output_tensor_arr inside the red box was the output of the model. I thought that the model result, obj->tidlObj.output_tensor_arr[0], obj->tidlObj.output_tensor_arr[1], could be put as each post-processing input. But the graph is not created properly on app_create_graph_tidl side even before post-processing starts.

If the model compiles properly on the pc, shouldn't app_create_graph_tidl also proceed successfully??? I'm not sure where the problem is.

thank you.

Hi,

Nikhil Dasan said:

You have already provided a part of the application log in the screenshot attached in the description, but it would be helpful if I get the full logs along with the remote side logs enabled too on TDA4VM

Could you share the logs along with the remote core logs, so that we could identify where is the error in the tidl node, that is causing failure during verify graph.

Regards,
Nikhil