TDA4VM: TDA4VM: Query regarding CenterNet model postprocessing

mudasar kalwar

Part Number: TDA4VM

Hi,

I have successfully imported centerNet model into tidl application and it is compiling.

Now I have to add postprocessing to see the output. output of centerNet are three convolution layers with 1x4x80x128, 1x2x80x128, 1x2x80x128 respectively.

As I can see tidl code output of centerNet is stored in vx_object_array output_tensor_arr

My question is

How can I retrieve the data from output_tensor_arr to any normal c/c++ array ?

I try to google but there is no luck finding anything useful.

Thank you

Regards,

Mudasar

over 3 years ago

0 Lucas Weaver over 3 years ago

TI__Genius 13845 points

Hello Mudasar,

A few questions first:

Which SDK version are you using?

Are you using a custom TIDL application or are you referencing an existing application below from vision_apps?

https://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/latest/exports/docs/vision_apps/docs/user_guide/DL_DEMOS.html

And for the post processing, are you using a custom node or using an existing vision_apps node?

Regards,

Lucas

0 mudasar kalwar over 3 years ago in reply to Lucas Weaver

Prodigy 170 points

Hello Lucas,

I am using SDK "ti-processor-sdk-rtos-j721e-evm-07_01_00_11".

I am refereeing to tidl_od application and for now which is already available in vision_apps, dl_demos.

for post processing for now I used same as in vision_apps nodes.

I am trying to get the prediction output from the CenterNet which output is 3 convolution layers of size 4x80x128, 2x80x128, 2x80x128.

So my goal is to get those outputs and postprocess them using normal c/c++ language or if possible using tiovx nodes as I need sigmoid, maxpooling and affine transform functions.

Please guide me how to implement above post process. and in case you have guide to create custom application it will be good also.

Thank you

Mudasar

0 mudasar kalwar over 3 years ago in reply to Lucas Weaver

Prodigy 170 points

Hello Lucas,

I am using SDK "ti-processor-sdk-rtos-j721e-evm-07_01_00_11".

I am refereeing to tidl_od application and for now which is already available in vision_apps, dl_demos.

for post processing for now I used same as in vision_apps nodes.

I am trying to get the prediction output from the CenterNet which output is 3 convolution layers of size 4x80x128, 2x80x128, 2x80x128.

So my goal is to get those outputs and postprocess them using normal c/c++ language or if possible using tiovx nodes as I need sigmoid, maxpooling and affine transform functions.

Please guide me how to implement above post process. and in case you have guide to create custom application it will be good also.

Thank you

Mudasar

Processors

Processors forum

TDA4VM: TDA4VM: Query regarding CenterNet model postprocessing