how to implement DCNv2_TRT plugin in onnx-tensorrt

[zengjie617789]

thank you for the awesome implement firstly. I compile the source code successfully and it can find the plugin in exporting onnx model, but i got the error about the output dim of DCNv2, and i think the reason relate to builtin_op_importers.cpp. I found there are some differences compared with other version DCNv2 plugin code, such as

void DCNv2PluginDynamic::configurePlugin(const DynamicPluginTensorDesc* inputs, int nbInputs,
	const DynamicPluginTensorDesc* outputs, int nbOutputs)
{
	// Validate input arguments
	assert(nbInputs == 4);
	assert(nbOutputs == 1);

Could you explain what the four inputs mean ? Here is my netron imageshoot about the input attributes:

thank you in advance.

[gpzhang7]

我也遇到了这个问题，请问您解决了吗？

[zengjie617789]

这个作者的c++那部分代码没把输入说明，可以参考这个： https://github.com/lesliejackson/TensorRT-DCNv2-Plugin

I cannot run the code without clarying the input node in C++. I refered the code below: https://github.com/lesliejackson/TensorRT-DCNv2-Plugin

[gpzhang7]

谢谢

[gpzhang7]

这个作者的c++那部分代码没把输入说明，可以参考这个： https://github.com/lesliejackson/TensorRT-DCNv2-Plugin

I cannot run the code without clarying the input node in C++. I refered the code below: https://github.com/lesliejackson/TensorRT-DCNv2-Plugin

您好，这个链接中好像需要编写InferPlugin.cpp文件，但是master中没找到这个文件，请问是否存在这个问题呢？

[zengjie617789]

作者没有提到需要InferPlugin.cpp文件呢。你可以把链接和需要的地方附件一下。

[gpzhang7]

作者没有提到需要InferPlugin.cpp文件呢。你可以把链接和需要的地方附件一下。

我搞出来了，谢了啊