J721S2XSOMXEVM: [J721S2][SDK] Object detection models in SDK

Soorya Prakash Sakthivel

Part Number: J721S2XSOMXEVM
Other Parts Discussed in Thread: TDA4VE-Q1

Hi Team,

We have run object detection models on our own custom video using gstreamer pipeline where objects are labeled as human, car, truck etc on TDA4VM-SK (Starter Kit) SDK by updating edgeai-gui-app on top of linux sdk.

We are trying to achieve the same on TDA4VE-Q1 EVM (J721S2).

While running tidl_od demos of J721S2 SDK, we can see the object bounding box but the objects are not labeled like in SK demos. Is there any configuration that needs to be done inorder to achieve the labeling part. ?

Also we want to know how to run models on custom video file in TDA4VE-Q1 EVM on top of SDK.

Thanks in advance,

Regards,

Soorya

3 months ago

0 Vaibhav Kumar 3 months ago

TI__Intellectual 2900 points

Hi,

For labels, you need to apply the attached patch on vision_apps/apps/dl_demos/app_tidl_od . https://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/791/app_5F00_tidl_5F00_od_5F00_changes.patch

For custom video file, you can refer to vision_apps/apps/basic_demos/app_multi_cam_codec. software-dl.ti.com/.../group_apps_basic_demos_app_multi_cam_codec.html

Regards,
Vaibhav

0 Soorya Prakash Sakthivel 3 months ago in reply to Vaibhav Kumar

Prodigy 80 points

Hi Vaibhav,

Thanks for the response, I was able to apply patch and could see labeling in PC Emulation.

app_multi_cam_codec supports just file read and display but is there any patch that makes it support running object detection models on top of read h264 file and label objects with bounding box. That is our expected usecase.

Thanks & Regards,

Soorya

0 Brijesh Jadav 3 months ago in reply to Soorya Prakash Sakthivel

TI__Guru**** 504675 points

Soorya,

no, multi-cam-codec examples is not integrated with TIDL.

Regards,

Brijesh

Processors

Processors forum

J721S2XSOMXEVM: [J721S2][SDK] Object detection models in SDK