IPluginV2 not found in sdk TensorRT5.0

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

I read the guide of tensorRT5.0 from web. I need to migrate Plugins from TensorRT 4.0.1 to TensorRT 5.0.0 RC. And I download the 5.0 tar from web. But I did not find any class named IPluginV2 in the headers which mentioned in the pdf. Where is it ?
By the way, I found the onnx-tensorrt is not the newest , even the header is not the same which made me a mistake in build new nvonnxparser.so to replace. I thought it should be some notice here. Thanks.

One more question. I used the TR 4.0.1.6 uff parser to parse the uff file OK, but in TR 5.0 which made me wrong. Returning the error message: Concat operation axis is out of bounds for layer. What is the wrong may it be? And how could I solve it?

2

Hello,

Regarding IPluginV2, it is described https://docs.nvidia.com/deeplearning/sdk/tensorrt-release-notes/rel_5-0-RC.html, and is described as available as GA feature, not in the current RC release.

3

I don’t understand your comment about onnx-tensorrt is not the newest? can you elaborate?

4

Hello, NVES. Thanks for your reply to the IPluginV2.
For Onnx-tensorrt mentioned above, I first build my project with the libnvonnxparser.so in the tar, but it is old that some ops are not supported. So I build the onnx-lib by project: GitHub - onnx/onnx-tensorrt: ONNX-TensorRT: TensorRT backend for ONNX which you posted on web. I replace the new libnvonnxparser* on the old lib. But actually the api in header is not the same. So I must change the functions which I used before , or else get a wrong result.

5

Hi,

NvOnnxParser.h in both TRT4 and TRT5 are the same vesion.

#define NV_ONNX_PARSER_MAJOR 0
#define NV_ONNX_PARSER_MINOR 1
#define NV_ONNX_PARSER_PATCH 0

What ops are no longer supported when you migrated to trt5?

6

Looking at the onnx-lib github, it looks to be TRT3-based.

7

Hi, such as Squeeze op which supported in Git’s.
The header version you posted are the same. But the TR5’s header has func “parseFromFile” while the Git’s not.And the func one in TR5’s is createParser(nvinfer1::INetworkDefinition& network,nvinfer1::ILogger& logger). The param “network” is reference, while in Git’s is pointer.

8

I want to know why my uff file can not work any more. I still used the factory to build plugin as before and adding some override needed funcs in plugin. Must I change to IPluginCreator ?

9

The uffparser error: I Changed to ‘IPuginCreator’ which still can not work. At last, I found the solution here https://devtalk.nvidia.com/default/topic/1042385/tensorrt/tensorrt-5-0rc-on-xavier-jetpack-4-0ea-vs-tensorrt-4-0ga-on-tx2-jetpack-3-3-/?offset=5#5289402.