Can the default InstanceNormalization replacement with MMCVInstanceNormalization be made optional?

I’m with mmcv-full==1.3.16, trying to export trt engine with mmcv’s exporter but failed with,

[06/07/2022-00:28:05] [TRT] [E] 3: getPluginCreator could not find plugin: MMCVInstanceNormalization version: 1
Traceback (most recent call last):
  File "tools/deployment/onnx2tensorrt.py", line 145, in <module>
    workspace_size=args.workspace_size)
  File "tools/deployment/onnx2tensorrt.py", line 47, in onnx2tensorrt
    max_workspace_size=max_workspace_size)
  File "/path/to/my/conda_env/open-mmlab/lib/python3.7/site-packages/mmcv/tensorrt/tensorrt_utils.py", line 63, in onnx2trt
    raise RuntimeError(f'parse onnx failed:\n{error_msgs}')
RuntimeError: parse onnx failed:
In node 6 (importFallbackPluginImporter): UNSUPPORTED_NODE: Assertion failed: creator && "Plugin not found, are the plugin name, version, and namespace correct?"

My onnx can be successfully exported by native trtexec. It seems TensorRT 7.2+ readily supports this operation. I have to otherwise register MMCVInstanceNormalization plugin somehow. Could you also show me show to do the latter?

[grimoire]

We have move TensorRT support to MMDeploy. Which supports InstanceNorm with TensorRT>=8.0

[serser]

This however generated an engine that can’t be deserialized due to tag mismatch. Confirmed to have same version of TensorRT.

As far as I know, the tag is used to distinguish different TensorRT versions. Could you check the environment again? You can use ldd on the related libraries to see which TensorRT are they linked to.

Aye. I was using the version compiled w/ the above mentioned flags to convert engines (mmcv.tensorrt.onnx2trt) and loaded it with mmcv.tensorrt.TRTWraper, all along with TensorRT-7.2.3.4 python package. I am puzzled to find the version mismatch though.

My bad, not loading correct engine file. But a new error shows up,

[06/07/2022-21:50:05] [TRT] [E] 1: [pluginV2Runner.cpp::load::290] Error Code 1: Serialization (Serialization assertion creator failed.Cannot deserialize plugin since corresponding IPluginCreator not found in Plugin Registry)
[06/07/2022-21:50:05] [TRT] [E] 4: [runtime.cpp::deserializeCudaEngine::50] Error Code 4: Internal Error (Engine deserialization failed.)

Solved with trt.init_libnvinfer_plugins(None, "") mentioned here.