Static Symbolic Method Operator

I found a piece of code that adds TRT::EfficientNMS_TRT to the YOLOv8 network and exports it to ONNX. The TRT::EfficientNMS_TRT used in it seems to be similar to efficientNMSPlugin, but its parameters are slightly different from those given in the documentation, with the addition of suffixes(e.g. _f) representing data types. In this code, the forward function only returns random tensors. I would like to know the specific requirements or limitations of the forward function and symbolic function (are the input parameters the same?). The PyTorch documentation only provides information about TorchScript operators and ONNX standard operators, but I want to know how TensorRT plugins are supported in g.op(). Where can I find tutorials or documentation related to this?

1 Like