DaChao_Xu

> @deep-practice > 7.0.0.11 use opset=10 when export onnx models onnx版本有什么要求吗？

> 导出ONNX模型时，试一试opset=10，7,0.0.11及以下不支持opset=12 请问tensorrt版本和支持的Opset具体在哪里看呢？

> @deep-practice > 7.0.0.11 use opset=10 when export onnx models is right!!

我试了https://github.com/linghu8812/tensorrt_inference/tree/master/Yolov4 的int8推理，结果是正确的，只是精度不够。最近看了一些TRT的int8量化资料，精度下降太多可能与训练时使用的激活函数有关，猜测是TRT只对正半轴做了8bits quantize，yolov4使用了mish激活函数，所以feature map是存在负值的，yolov4可以使用ReLU替换Leaky-ReLU，我认为这点替换后的精度损失相比于int8的精度下降是可以接受的。还有一点思考就是如果TRT真的只做正半轴的话，sigmoid激活函数可能需要在解析层来做了。所以yolov5 int8应该没问题。

/usr/local/include/NvOnnxParser.h:27:34: fatal error: NvOnnxParserTypedefs.h: 没有那个文件或目录 compilation terminated. CMakeFiles/Yolov4_trt.dir/build.make:86: recipe for target 'CMakeFiles/Yolov4_trt.dir/Yolov4.cpp.o' failed make[2]: *** [CMakeFiles/Yolov4_trt.dir/Yolov4.cpp.o] Error 1 CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/Yolov4_trt.dir/all' failed make[1]: *** [CMakeFiles/Yolov4_trt.dir/all] Error 2 Makefile:127:...

@linghu8812 Input filename: ../cfg/yolov4-csp.onnx ONNX IR version: 0.0.5 Opset version: 10 Producer name: darknet to ONNX example Producer version: Domain: Model version: 0 Doc string: ---------------------------------------------------------------- WARNING: ONNX model has...

tensorrt-6.1.0.5

@linghu8812 ,hello,can you upload a yolov4-csp.onnx?

@linghu8812 hello，use your onnx，the result is right, i think maybe convert onnx wrong, i use python3.6 + onnx1.5.0, the results is terrible！ 

@linghu8812 hello，i used the last version，the results no change，is wrong。 ./Yolov4_trt ../config-xmish.yaml ../samples/ ---------------------------------------------------------------- Input filename: ../cfg/yolov4x-mish-normal-best.onnx ONNX IR version: 0.0.5 Opset version: 10 Producer name: darknet to ONNX example...