squeezenet to trt 7

squeezenet/CMakeLists.txt

@@ -9,8 +9,13 @@ set(CMAKE_CXX_STANDARD 11)
 set(CMAKE_BUILD_TYPE Debug)
 
 include_directories(${PROJECT_SOURCE_DIR}/include)
-include_directories(/usr/local/cuda-9.0/targets/aarch64-linux/include)
-link_directories(/usr/local/cuda-9.0/targets/aarch64-linux/lib)
+# include and link dirs of cuda and tensorrt, you need adapt them if yours are different
+# cuda
+include_directories(/usr/local/cuda/include)
+link_directories(/usr/local/cuda/lib64)
+# tensorrt
+include_directories(/usr/include/x86_64-linux-gnu/)
+link_directories(/usr/lib/x86_64-linux-gnu/)
 
 add_executable(squeezenet ${PROJECT_SOURCE_DIR}/squeezenet.cpp)
 target_link_libraries(squeezenet nvinfer)

squeezenet/README.md

@@ -8,11 +8,6 @@ SqueezeNet 1.1 has 2.4x less computation and slightly fewer parameters
 
 For the Pytorch implementation, you can refer to [pytorchx/squeezenet](https://github.com/wang-xinyu/pytorchx/tree/master/squeezenet)
 
-Following tricks are used in this squeezenet,
-
-- MaxPool2d(ceil_mode=True), ceilmode=True, which is not supported in Tensorrt4, we use a padding layer before maxpool to solve this problem.
-- For Pytorch AdaptiveAvgPool2d(), we use fixed input dimension, and use regular average pooling to replace it.
-
 ```
 // 1. generate squeezenet.wts from [pytorchx/squeezenet](https://github.com/wang-xinyu/pytorchx/tree/master/squeezenet)
 
@@ -36,4 +31,3 @@ sudo ./squeezenet -d   // deserialize plan file and run inference
 // 4. see if the output is same as pytorchx/squeezenet
 ```
 
-