Object detection using yolo (with preprocessing, postprocessing) tutorial


I have a Nvidia jetson orin NX and I decided to give TensorRT a try. The solution right now is working with opencv only. The idea of project is to process frames using yolo for object detection following. To do this I do the following:

  • Capture a frame from a 4K camera

  • Preprocess the frame so that it is resized to a specific width and height. I’m assuming a .engine/.trt model has a specific input size. I converted onnx (1280, 1280) to .trt.

  • I need to postprocess the detections. I need to have the bounding boxes coordinates, confidences, etc.

I searched on the internet for this specific scenario, and I couldn’t find any recent tutorials that could give me an insight. I found some tutorials but they are outdated (before TensorRT 8.0.0), so the code functions are deprecated and removed. Also, the majority of the tutorials are for streams which is not what I need.
Taking into account what I said above, do you know any tutorial that could help me?


1 Like