PROCESSOR-SDK-AM68A: Artifact generation for facial landmark detection task

Hi,

Sadat Hossain said:

I am using TI Evaluation module (PROCESSOR-SDK-AM68A) to deploy Advanced driver-assistance system models. I have generated artifacts and deployed the Human Pose Estimation model (Yolox-s pose) to the board, it works well.

Great to hear this !

Sadat Hossain said:

here is no pipeline given by Edge-AI benchmark repository.

You can use our edgeai-tidl-tools repo to understand more about model compilation and inference.

Here is the link : https://github.com/TexasInstruments/edgeai-tidl-tools

You can look at OSRT Python based examples to get started on custom model compilation.

--

Pratik

Closing the thread.

Hi,

Please OSRT examples are available in edgeai tidl tools repos, the tools repo link is shared above.

Sharing the specific link for your ref : https://github.com/TexasInstruments/edgeai-tidl-tools/blob/master/examples/osrt_python/README.md

Currently Key point det is supported as part of model maker, we will update on facial landmark det in near future, if its in pipeline.

--

Pratik

Sure,

We recommend to create new thread, if you have any issues with your experiments.

Thanks

Pratik

Hi

Please checkout the documentation on the same here : 

https://github.com/TexasInstruments/edgeai-gst-apps-human-pose

Que 1 : 

rj z said:

when i import onnx and prototxt with this link https://github.com/TexasInstruments/edgeai-yolov5/tree/master/pretrained_models/models/keypoint/coco/edgeai-yolov5 to bin file,  show me error below.

【

more than one value provided for SINGLE parameter 'modelType' - picking up only first value

Parameter value parse for 'inDataFormat' = 'inWidth'(Conversion error)...

Please shared detailed errors logs, your experiment setup, executed command whenever posting question related to error issues.

Could you please create new thread for the same as this is slightly deviating from current E2E title, thank you.

Que 2 : 

rj z said:

how could i use keypoints model such as edgeai-yolo-pose or edgeai-yolox to import to generate bin file, then deploy keypoints model on TDA4??

Please look at our edgeai tidl tools repos, to understand more on how to compile custom models for TI SoCs

https://github.com/TexasInstruments/edgeai-tidl-tools

Que 3 : 

rj z said:

and does  https://github.com/TexasInstruments/edgeai-gst-apps-human-pose support TDA4VM board?

Lets understand how to check the github documentation, to help you get answer for this question.

On the github demo link shared above https://github.com/TexasInstruments/edgeai-gst-apps-human-pose

If you can scroll down the readme file, you will see section called Supported Devices adding direct link for your reference https://github.com/TexasInstruments/edgeai-gst-apps-human-pose#supported-devices