Deepstream integration for Yolo-pose

[KushalHM]

Hi,
Great work with the repo, Thank you for that. I was able to generate a yolo pose fp16 engine using this repo and get it to work on my GPU (RTX 2070).
Have you tried or are you aware of using the generated engine with nvinfer plugin of deepstream? I want to use it in deepstream 6.1.1 (with Cuda 11.7).

[MariJothi]

can you tell me, what version of tensorrt you using? i am having issue on creating tensorrt engine.

[aurelm95]

I am also interested in deploying the key point model with deepstream. Any update on this?

[KushalHM]

can you tell me, what version of tensorrt you using? i am having issue on creating tensorrt engine.

Hi, the tensorrt version is 8.5.2.2 but I had it also working in tensorrt 8.4.1. One thing to note, depending on the cuda/tensorrt version, the node numbers in the onnx file will change in add_custom_yolo_op.py. Please use netron to visualize the onnx model and use the correct numbers for inputs.

[KushalHM]

I am also interested in deploying the key point model with deepstream. Any update on this?

No update yet. I am keeping an eye on this repo also to see if there is any update, https://github.com/marcoslucianops/DeepStream-Yolo.git.
Might have to write the custom yolov7-pose deepstream inference + post processing plugin myself.

[MarcoPrassel]

I am trying to integrate the model yolov7-w6-pose-sim-yolo-fp16.engine on Deepstream but I get this error

[aurelm95]

I am interested in learning how to develop the neccessary plugins ando/or parsers for yolo pose deepstream. How could I start ? is there any documentation for that?

btw, @MarcoPrassel , could you share your progress with the code?

[YunghuiHsu]

Hi
I used yolov8-pose (which supports dynamic batch processing) instead and integrated it into the Deepstream Python API.
Detailed in deepstream-yolo-pose