DeepStream Gst-nvinfer standard format of outputs of detection models

I’m trying to use a custom object detection model on DeepStream.
The model is automatically converted to TensorRT engine file and ran just fine.

The problem is I don’t know how the default bbox parsing function of DeepStream looks like, so I have to write a custom bbox parsing function to parse my model’s outputs.

I want to know what is the standard format for object detection model’s outputs (confident score and bounding box regression) such as the meaning of tensor values, the tensor size, so when I create a custom model and use it with DeepStream, I don’t have to create my own bboxing parsing function.

Is there any documentation about that standard format?

Your model probably was somehow evaluated on python. You should port this python code to C++.