Hello, I’m kinda new to computer vision. I started a personal project that is the following.
I have a Jetson TX2 NX and a camera plugged on it. My YoloV7 model is already trained and has a really decent FPS detection on my computer.
My goals are: 1) to perform object detection in real time with YOLOV7.
2) Let the choice to the operator that sees the screen (on a computer) in real time, to choose only one of the object detected to track it. Hence it then becomes a single object tracking task.
At first I thaught about use YOLOV7 and OpenCV in Python, but after few research I found out that convert the YOLOV7 model to ONNX to TensorRT was the best option for fast inference. And I also discovered the Deepstream SDK that include things like VPI. I also found a repo call DeepStream-YOLO but I’m not sure how to use it.
Some new questions came out then: -Am I able to implement 2) without OpenCV and only with things like VPI through the Deepstream SDK ?
-As I know how to code in both Python and C++ what should be the best option to perform the real time object tracking task ?
-I know that I can run inference with TensorRT on the Jetson TX2 NX but to what extend can I also use DeepstreamSDK ? Will I encounter some restriction compared to the usage on a classic computer ?
Thank you in advance for any kind of help / informations, as I just begin into this world.