DCNV2 plugin tensorrt 8.5 arm int8 convert error

[mairkiss]

hello, i asked the same question before,i am using tensorrt8.5 to convert my onnx model to int8 model, the special thing is that my onnx model contains a custom DCNV2 plugin, The python implementation of DCNV2 is based on the following library： https://github.com/CharlesShang/DCNv2/blob/master/dcn_v2.py

But when converting the model, I always get such an error： ....... [2023-01-31 06:14:12 WARNING] Missing scale and zero-point for tensor 728, expect fall back to non-int8 implementation for any layer consuming or producing given tensor [2023-01-31 06:14:12 WARNING] Detected invalid timing cache, setup a local cache instead [2023-01-31 06:14:19 WARNING] No implementation of layer Split_120[SplitPlugin] obeys the requested constraints in strict mode. No conforming implementation was found i.e. requested layer computation precision and output precision types are ignored, using the fastest implementation. [2023-01-31 06:14:20 ERROR] 2: [pluginV2Runner.cpp::getInputHostScale::88] Error Code 2: Internal Error (Assertion scales.size() == 1 failed.) terminate called after throwing an instance of 'std::runtime_error' what(): Failed to create object

The previous suggestion was to update the version of tensorrt to 8.5, so I used the latest version below to replace： I download two 8.5 versions, x86 and arm, x86 can convert success, but arm failed， I don't know if there is a problem with my arm version download？i use cuda 11.4， platform: orin driver

[zerollzeng]

Looks like your model is building with INT8 but your plugin doesn't have the INT8 scale.

[zerollzeng]

Can you provide a reproduce for this error? I can help take a check if I got spare time.

[ttyio]

closing since no activity for more than 3 weeks. thanks!