Update README.md,windows.md and demo.cpp

Small fixes in DeformableConvRT.cpp

CMakeLists.txt

@@ -3,10 +3,12 @@ project(tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 set(CMAKE_CXX_STANDARD 14)
 
+option(ENABLE_OPENCV_CUDA_CONTRIB "Enable OpenCV CUDA Contrib" OFF )
+
 find_package(CUDA 9.0 REQUIRED)
 if (CUDA_FOUND)
-    #Get CUDA compute capability
     set(OUTPUTFILE ${CMAKE_CURRENT_SOURCE_DIR}/cmake/cuda_script) # No suffix required
+    execute_process(COMMAND "rm ${OUTPUTFILE}")
     set(CUDAFILE ${CMAKE_CURRENT_SOURCE_DIR}/cmake/getCudaArch.cu)
     execute_process(COMMAND ${CUDA_NVCC_EXECUTABLE} -lcuda ${CUDAFILE} -o ${OUTPUTFILE})
     execute_process(COMMAND ${OUTPUTFILE}
@@ -25,10 +27,7 @@ if (CUDA_FOUND)
         message(STATUS "CUDA Path: ${CUDA_TOOLKIT_ROOT_DIR}")
         message(STATUS "CUDA Libararies: ${CUDA_LIBRARIES}")
         message(STATUS "CUDA Performance Primitives: ${CUDA_npp_LIBRARY}")
-
         set(CUDA_NVCC_FLAGS "${ARCH}")
-        add_definitions(-DGPU) #You may not require this
-
     else()
         message(WARNING ${ARCH})
     endif()
@@ -51,7 +50,7 @@ endif()
 
 if(WIN32)
     if(CMAKE_BUILD_TYPE MATCHES Release)
-        set(CMAKE_CXX_FLAGS "/O2 /FS /EHsc /Md")
+        set(CMAKE_CXX_FLAGS "/O2 /FS /EHsc /MD")
         set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
     endif()
 
@@ -105,6 +104,13 @@ include_directories(${EIGEN3_INCLUDE_DIR})
 
 find_package(OpenCV REQUIRED)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DOPENCV")
+if(ENABLE_OPENCV_CUDA_CONTRIB)
+    if (OpenCV_FOUND)
+        find_package(OpenCV COMPONENTS cudawarping cudaarithm)
+        add_compile_definitions(OPENCV_CUDACONTRIB)
+        message("OpenCV Cuda Contrib modules found")
+    endif()
+endif()
 # if(OpenCV_CUDA_VERSION)
 #     add_compile_definitions(OPENCV_CUDACONTRIB)
 # endif()

README.md

@@ -105,13 +105,14 @@ When using openCV not compiled with contrib, comment the definition of OPENCV_CU
 
 ## How to compile this repo
 Build with cmake. If using Ubuntu 18.04 a new version of cmake is needed (3.15 or above).
+On both linux and windows ,the ```CMAKE_BUILD_TYPE``` variable needs to be defined as either ```Release``` or ```Debug```.
 ```
 git clone https://github.com/ceccocats/tkDNN
 cd tkDNN
 mkdir build
 cd build
-cmake .. 
+cmake -DCMAKE_BUILD_TYPE=Release -G"Ninja" .. 
-make
+ninja
 ```
 
 ## Workflow
@@ -136,7 +137,7 @@ For specific details on how to run:
 
 ![demo](https://user-images.githubusercontent.com/11562617/72547657-540e7800-388d-11ea-83c6-49dfea2a0607.gif)
 
-## tkDNN on Windows 10 (experimental)
+## tkDNN on Windows 10/11 (experimental)
 
 For specific details on how to run tkDNN on Windows 10 see [HERE](./docs/windows.md).
 

demo/demo/demo.cpp

@@ -45,16 +45,16 @@ int main(int argc, char *argv[]) {
 
     char ntype = 'y';
     if(argc > 2)
-        ntype = argv[2][0];
+        input = argv[2];
-    int n_classes = 80;
     if(argc > 3)
-        n_classes = atoi(argv[3]);
+        ntype = argv[3][0];
+    int n_classes = 80;
     if(argc > 4)
-        cfgPath = argv[4];
+        n_classes = atoi(argv[4]);
     if(argc > 5)
-        namePath = argv[5];
+        cfgPath = argv[5];
     if(argc > 6)
-        input = argv[6];
+        namePath = argv[6];
     int n_batch = 1;
     if(argc > 7)
         n_batch = atoi(argv[7]);

docs/demo.md

@@ -26,17 +26,23 @@ rm yolo4_fp32.rt        # be sure to delete(or move) old tensorRT files
 ```
 If you get problems in the creation, try to check the error activating the debug of TensorRT in this way:
 ```
-cmake .. -DDEBUG=True
+cmake .. -DCMAKE_BUILD_TYPE=Debug -DDEBUG=True
 make
 ```
 
-Once you have successfully created your rt file, run the demo: 
+Once you have successfully created your rt file, run the demo(yolo) :
 ```
-./demo yolo4_fp32.rt ../demo/yolo_test.mp4 y
+./demo yolo4_fp32.rt ../demo/yolo_test.mp4 y 80 ../tests/darknet/cfg/yolo4.cfg ../tests/darknet/names/cococ.names 
 ```
+
+To run demo for mobilenet and centernet for the created rt file :
+```
+./demo mobilenetv2ssd_fp32.rt m 20
+```
+
 In general the demo program takes 7 parameters:
 ```
-./demo <network-rt-file> <path-to-video> <kind-of-network> <number-of-classes> <n-batches> <show-flag> <conf-thresh>
+./demo <network-rt-file> <path-to-video> <kind-of-network> <number-of-classes> <cfg-path> <name-path>  <n-batches> <show-flag> <conf-thresh>
 ```
 where
 
@@ -44,6 +50,8 @@ where
 * ```<<path-to-video>``` is the path to a video file or a camera input  
 * ```<kind-of-network>``` is the type of network. Thee types are currently supported: ```y``` (YOLO family), ```c``` (CenterNet family) and ```m``` (MobileNet-SSD family)
 * ```<number-of-classes>```is the number of classes the network is trained on
+* ```<cfg-path> ```is the relative path to the config file (only for darknet based networks) used to train the network
+* ```<name-path>```is the relative path to the names file (only for darknet based networks) used to train the network
 * ```<n-batches>``` number of batches to use in inference (N.B. you should first export TKDNN_BATCHSIZE to the required n_batches and create again the rt file for the network).
 * ```<show-flag>``` if set to 0 the demo will not show the visualization but save the video into result.mp4 (if n-batches ==1)
 * ```<conf-thresh>``` confidence threshold for the detector. Only bounding boxes with threshold greater than conf-thresh will be displayed.

docs/windows.md

@@ -7,17 +7,18 @@
  - [Run the demo on Windows](#run-the-demo-on-windows)
     - [FP16 inference windows](#fp16-inference-windows)
     - [INT8 inference windows](#int8-inference-windows)
+ - [Run tkDNN on WSL2 with cuda](#tkdnn-on-cuda-wsl)
  - [Known issues with tkDNN on Windows](#known-issues-with-tkdnn-on-windows)
 
 ### Dependencies-Windows 
 This branch should work on every NVIDIA GPU supported in windows with the following dependencies:
 
 * WINDOWS 10 1803 or HIGHER 
-* CUDA 10.0 (Recommended CUDA 11.2 )
+* CUDA 11.2 
-* CUDNN 7.6 (Recommended CUDNN 8.1.1 )
+* CUDNN 8.1.1
-* TENSORRT 6.0.1 (Recommended TENSORRT 7.2.3.4 )
+* TENSORRT 7.2.3 
-* OPENCV 3.4 (Recommended OPENCV 4.2.0 )
+* OPENCV 4.2  
-* MSVC 16.7 
+* MSVC 16.9+
 * YAML-CPP 
 * EIGEN3
 * 7ZIP (ADD TO PATH)
@@ -58,7 +59,7 @@ To run the object detection file create .rt file bu running:
 
 Once the rt file has been successfully create,run the demo using the following command:
 ```
-.\demo.exe yolo4tiny_fp32.rt ..\demo\yolo_test.mp4 y 
+.\demo.exe yolo4_fp32.rt ..\demo\yolo_test.mp4 y 80 ..\tests\darknet\cfg\yolo4.cfg ..\tests\darknet\names\cococ.names 
 ```
  For general info on more demo paramters,check Run the demo section on top 
  To run the test_all_tests.sh on windows,use git bash or msys2 
@@ -85,11 +86,17 @@ del /f  yolo4tiny_int8.rt        # be sure to delete(or move) old tensorRT files
 
 ```
 
+### Run tkDNN on WSL2 with cuda
+tkDNN works on wsl2 with cuda,although not all networks (centernet,mobilenet) work properly.
+If you encounter issues with running the network as a result of driver not found or cuda launch error,running the following command should solve the issue
+```cp /usr/lib/wsl/lib/lib* /usr/lib/x86_64-linux-gnu/ ```
+
+
+
 ### Known issues with tkDNN on Windows
 
-Mobilenet and Centernet demos work properly only when built with msvc 16.7 in Release Mode,when built in debug mode for the mentioned networks one might encounter opencv assert errors
+In theory all models (centernet,mobilenet,darknet,centertrack,cnet3d and shelfnet) should work on Windows.
 
-All Darknet models work properly with demo using MSVC version(16.7-16.9)
+On pascal cards(sm 6x) ,nvidia cuda wsl driver 510.06 don't work well with tkDNN both on windows and cuda wsl , Nvidia drivers >465+ and < 500 are completely supported .
 
-It is recommended to use Nvidia Driver(465+),Cuda unknown errors have been observed when using older drivers on pascal(SM 61) devices.
 

src/pluginsRT/DeformableConvRT.cpp

@@ -65,7 +65,6 @@ DeformableConvRT::~DeformableConvRT() {
     checkCuda( cudaFree(offset) );
     checkCuda( cudaFree(mask) );
     checkCuda( cudaFree(ones_d2) );
-    cublasDestroy(handle);
 }
 
 DeformableConvRT::DeformableConvRT(const void *data, size_t length) {
@@ -229,7 +228,9 @@ void DeformableConvRT::serialize(void *buffer) const NOEXCEPT {
     assert(buf == a + getSerializationSize());
 }
 
-void DeformableConvRT::destroy() NOEXCEPT { delete this; }
+void DeformableConvRT::destroy() NOEXCEPT {
+    delete this;
+}
 
 
 const char *DeformableConvRT::getPluginNamespace() const NOEXCEPT {