system
January 8, 2020, 8:51am
1
In the samples I see the sampleFasterRCNN with caffe parser with plugin, and sampleUffFasterRCNN with uff parser.
And There is IpluginFactory in both NvCaffeParser.h and NvUffParser.h, which is not found in NvOnnxParser.h
So does the onnx parser not yet suppport models require plugins now in trt7? Or the onnx models do not need plugin?
Thanks
Hi,
You convert your model to ONNX using tf2onnx and then convert to TensorRT using ONNX parser.
Any layer that are not supported can be replaced by custom plugin in ONNX parser as well.
Thanks
Is there anyway to add plugins to your ONNX file if your model is coming from PyTorch?
Hi,
You can use pytorch to ONNX converter.
Please refer to below link for more details to add custom plugin:
https://github.com/NVIDIA/TensorRT/issues/6#issuecomment-570290621
Thanks