Face landmarks model not working on deepstream-test2 app

[muhammadanas]

• Hardware Platform: Jetson Nano
• DeepStream Version: 6.0.1
• JetPack Version: 4.6.1 (valid for Jetson only)
• TensorRT Version: 8.2.1.8
• Python Version: 3.6.9
• CUDA Version: 10.2.300

Hi guys, I want some guidelines on the issue I am facing. I have loaded FaceNet model in primary while Facial landmarks model in secondary config file in the test-2 sample app. Here is the config settings for pgie file.

and this one for sgie.txt file

I am only getting predictions for the FaceNet model (i have also implemented tracker). Below is the output video.

As for the Facial landmarks model, the engine has been created but there are no inference logs showing up neither do I see landmarks on the faces. As for the record, I am using this config file settings from this github repo: GitHub - latonaio/facial-landmarks-estimation-on-deepstream

Also I have commented out the functionalities for sgie2 and sgie3 files and properly re-updated the code but still why I am not getting any inference?

What could be the main issue? Could it be with py file because I have already tried modifying config files several times, but still looking to guidelines from you guys on that.

As engine file for landmarks model created successfully, which maybe hints that model is loaded and working fine and maybe I am unable to get output metadata from the model?

Please feel free to ask any questions.

[Fiona.Chen]

Please refer to deepstream_tao_apps/apps/tao_others/deepstream-faciallandmark-app at release/tao3.0_ds6.0.1 · NVIDIA-AI-IOT/deepstream_tao_apps (github.com)

deepstream-app does not support such model.

[muhammadanas]

I have downloaded these facenet and facial landmark estimation models from Nvidia ngc catalog, also facenet model is working fine as a primary detector. the problem is with the facial landmarks model not getting output from the primary detector.

[Fiona.Chen]

Do you run the sample according to the “Build and Run” part in the README? deepstream_tao_apps/apps/tao_others/deepstream-faciallandmark-app at release/tao3.0_ds6.0.1 · NVIDIA-AI-IOT/deepstream_tao_apps (github.com)

[Fiona.Chen]

Since you are working with DeepStream 6.0.1, you can only get the “release/tao3.0_ds6.0.1” branch of NVIDIA-AI-IOT/deepstream_tao_apps at release/tao3.0_ds6.0.1 (github.com)

[muhammadanas]

Yeah, I now followed the “Bulild and Run” method in the “release/tao3.0_ds6.0.1” branch and i am only getting the no of frames and number of detected faces in the output log. Although the engine file has been created for facial landmarks model but still i am not getting any landmarks on the faces.

[Fiona.Chen]

Can you provide your test video?