ROS Noetic + YOLOV4 (Darknet by AlexeyAB) + CUDNN + CUDNN_HALF

.gitmodules

@@ -1,3 +1,3 @@
 [submodule "darknet"]
 	path = darknet
-	url = https://github.com/leggedrobotics/darknet
+	url = https://github.com/AlexeyAB/darknet.git

darknet_ros/CMakeLists.txt

@@ -6,35 +6,89 @@ set(CMAKE_CXX_STANDARD 11)
 set(CMAKE_C_FLAGS "-Wall -Wno-unused-result -Wno-unknown-pragmas -Wno-unused-variable -Wfatal-errors -fPIC ${CMAKE_C_FLAGS}")
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
+enable_language(C)
+enable_language(CXX)
+
 # Define path of darknet folder here.
 find_path(DARKNET_PATH
   NAMES "README.md"
   HINTS "${CMAKE_CURRENT_SOURCE_DIR}/../darknet/")
 message(STATUS "Darknet path dir = ${DARKNET_PATH}")
 add_definitions(-DDARKNET_FILE_PATH="${DARKNET_PATH}")
 
+#Darknet extra modules
+set(CMAKE_MODULE_PATH "${DARKNET_PATH}/cmake/Modules/" ${CMAKE_MODULE_PATH})
+
 # Find CUDA
 find_package(CUDA QUIET)
+
 if (CUDA_FOUND)
-  find_package(CUDA REQUIRED)
-  message(STATUS "CUDA Version: ${CUDA_VERSION_STRINGS}")
-  message(STATUS "CUDA Libararies: ${CUDA_LIBRARIES}")
-  set(
-    CUDA_NVCC_FLAGS
-    ${CUDA_NVCC_FLAGS};
-    -O3
-    -gencode arch=compute_30,code=sm_30
-    -gencode arch=compute_35,code=sm_35
-    -gencode arch=compute_50,code=[sm_50,compute_50]
-    -gencode arch=compute_52,code=[sm_52,compute_52]
-    -gencode arch=compute_61,code=sm_61
-    -gencode arch=compute_62,code=sm_62
-  )
-  add_definitions(-DGPU)
+  # Get CUDA Compute capability with modified script from internet
+  set(CUDAFILE ${CMAKE_CURRENT_SOURCE_DIR}/scripts/check_cuda.cu)
+  set(OUTPUTFILE ${CMAKE_CURRENT_SOURCE_DIR}/scripts/cuda_check) 
+  # Compile cuda arch check script
+  execute_process(COMMAND nvcc -lcuda ${CUDAFILE} -o ${OUTPUTFILE})
+  execute_process(COMMAND ${OUTPUTFILE}
+                  RESULT_VARIABLE CUDA_RETURN_CODE
+                  OUTPUT_VARIABLE GENCODE)
+  #Check correct architectures and capabilities
+  if(${CUDA_RETURN_CODE} EQUAL 0)
+    set(CUDA_SUCCESS "TRUE")
+  else()
+    message(Warning"${GENCODE}. Forcing darknet_ros without CUDA")
+    set(CUDA_SUCCESS "FALSE")
+  endif()
+  #remove cuda_check binary
+  execute_process(COMMAND rm -f ${OUTPUTFILE})
+  #Config CUDA
+  if (${CUDA_SUCCESS})
+    set(
+      CUDA_NVCC_FLAGS
+      ${CUDA_NVCC_FLAGS};
+      -O3
+      #-gencode arch=compute_30,code=sm_30  deprecated for CUDA9+
+      #-gencode arch=compute_35,code=sm_35  deprecated for CUDA9+
+      # -gencode arch=compute_50,code=[sm_50,compute_50]
+      # -gencode arch=compute_52,code=[sm_52,compute_52]
+      # -gencode arch=compute_61,code=sm_61
+      # -gencode arch=compute_62,code=sm_62
+      ${GENCODE}       # Force to compile to installed GPUs Architectures 
+    )
+    message(STATUS "CUDA Architecture: ${GENCODE}")
+    message(STATUS "CUDA Version: ${CUDA_VERSION_STRING}")
+    message(STATUS "CUDA Version: ${CUDA_VERSION_STRINGS}")
+    message(STATUS "CUDA Libararies: ${CUDA_LIBRARIES}")
+    add_definitions(-DGPU)
+  else()
+    list(APPEND LIBRARIES "m")
+  endif()
 else()
   list(APPEND LIBRARIES "m")
 endif()
 
+# Find CUDNN
+if(CUDA_SUCCESS)
+  find_package(CUDNN)
+  if (CUDNN_FOUND)
+    message(STATUS "CUDNN Version: ${CUDNN_VERSION_STRINGS}")
+    message(STATUS "CUDNN Libararies: ${CUDNN_LIBRARIES}")
+    set(ADDITIONAL_CXX_FLAGS "${ADDITIONAL_CXX_FLAGS} -DCUDNN")
+  # ENABLE CUDNN_HALF
+    if ( "-gencode arch=compute_70,code=sm_70" IN_LIST CUDA_NVCC_FLAGS OR
+        "-gencode arch=compute_72,code=sm_72" IN_LIST CUDA_NVCC_FLAGS OR
+        "-gencode arch=compute_75,code=sm_75" IN_LIST CUDA_NVCC_FLAGS OR
+        "-gencode arch=compute_80,code=sm_80" IN_LIST CUDA_NVCC_FLAGS OR
+        "-gencode arch=compute_86,code=sm_86" IN_LIST CUDA_NVCC_FLAGS)
+      set(ENABLE_CUDNN_HALF "TRUE" CACHE BOOL "Enable CUDNN Half precision" FORCE)
+      message(STATUS "Your setup supports half precision (CUDA_ARCHITECTURES >= 70)")
+    else()
+      set(ENABLE_CUDNN_HALF "FALSE" CACHE BOOL "Enable CUDNN Half precision" FORCE)
+      message(STATUS "Your setup does not support half precision (it requires CUDA_ARCHITECTURES >= 70)")
+    endif()
+  endif()
+endif()
+
+
 # Find X11
 message ( STATUS "Searching for X11..." )
 find_package ( X11 REQUIRED )
@@ -65,6 +119,15 @@ find_package(catkin REQUIRED
 add_definitions(-DOPENCV)
 add_definitions(-O4 -g)
 
+# Enable CUDNN in darknet
+if(CUDNN_FOUND)
+  add_definitions(-DCUDNN)
+  # Enable   in darknet
+  if(ENABLE_CUDNN_HALF)
+    add_definitions(-DCUDNN)
+  endif()
+endif()
+
 catkin_package(
   INCLUDE_DIRS
     include
@@ -86,60 +149,66 @@ catkin_package(
 include_directories(
   ${DARKNET_PATH}/src
   ${DARKNET_PATH}/include
+  ${DARKNET_PATH}/3rdparty/stb/include
   include
   ${Boost_INCLUDE_DIRS}
   ${catkin_INCLUDE_DIRS}
 )
 
 set(PROJECT_LIB_FILES
     src/YoloObjectDetector.cpp                    src/image_interface.cpp
+    src/yolo_object_detector_node.cpp
 )
 
 set(DARKNET_CORE_FILES
-    ${DARKNET_PATH}/src/activation_layer.c        ${DARKNET_PATH}/src/im2col.c
-    ${DARKNET_PATH}/src/activations.c             ${DARKNET_PATH}/src/image.c
-    ${DARKNET_PATH}/src/avgpool_layer.c           ${DARKNET_PATH}/src/layer.c
-    ${DARKNET_PATH}/src/batchnorm_layer.c         ${DARKNET_PATH}/src/list.c
-    ${DARKNET_PATH}/src/blas.c                    ${DARKNET_PATH}/src/local_layer.c
-    ${DARKNET_PATH}/src/box.c                     ${DARKNET_PATH}/src/lstm_layer.c
-    ${DARKNET_PATH}/src/col2im.c                  ${DARKNET_PATH}/src/matrix.c
-    ${DARKNET_PATH}/src/connected_layer.c         ${DARKNET_PATH}/src/maxpool_layer.c
-    ${DARKNET_PATH}/src/convolutional_layer.c     ${DARKNET_PATH}/src/network.c
-    ${DARKNET_PATH}/src/cost_layer.c              ${DARKNET_PATH}/src/normalization_layer.c
-    ${DARKNET_PATH}/src/crnn_layer.c              ${DARKNET_PATH}/src/option_list.c
-    ${DARKNET_PATH}/src/crop_layer.c              ${DARKNET_PATH}/src/parser.c
-    ${DARKNET_PATH}/src/cuda.c                    ${DARKNET_PATH}/src/region_layer.c
-    ${DARKNET_PATH}/src/data.c                    ${DARKNET_PATH}/src/reorg_layer.c
-    ${DARKNET_PATH}/src/deconvolutional_layer.c   ${DARKNET_PATH}/src/rnn_layer.c
-    ${DARKNET_PATH}/src/demo.c                    ${DARKNET_PATH}/src/route_layer.c
-    ${DARKNET_PATH}/src/detection_layer.c         ${DARKNET_PATH}/src/shortcut_layer.c
-    ${DARKNET_PATH}/src/dropout_layer.c           ${DARKNET_PATH}/src/softmax_layer.c
-    ${DARKNET_PATH}/src/gemm.c                    ${DARKNET_PATH}/src/tree.c
-    ${DARKNET_PATH}/src/gru_layer.c               ${DARKNET_PATH}/src/utils.c
-    ${DARKNET_PATH}/src/upsample_layer.c          ${DARKNET_PATH}/src/logistic_layer.c
-    ${DARKNET_PATH}/src/l2norm_layer.c            ${DARKNET_PATH}/src/yolo_layer.c
-    ${DARKNET_PATH}/src/iseg_layer.c              ${DARKNET_PATH}/src/image_opencv.cpp
-
-    ${DARKNET_PATH}/examples/art.c                ${DARKNET_PATH}/examples/lsd.c
-    ${DARKNET_PATH}/examples/nightmare.c          ${DARKNET_PATH}/examples/instance-segmenter.c
-    ${DARKNET_PATH}/examples/captcha.c            ${DARKNET_PATH}/examples/regressor.c
-    ${DARKNET_PATH}/examples/cifar.c              ${DARKNET_PATH}/examples/rnn.c
-    ${DARKNET_PATH}/examples/classifier.c         ${DARKNET_PATH}/examples/segmenter.c
-    ${DARKNET_PATH}/examples/coco.c               ${DARKNET_PATH}/examples/super.c
-    ${DARKNET_PATH}/examples/darknet.c            ${DARKNET_PATH}/examples/tag.c
-    ${DARKNET_PATH}/examples/detector.c           ${DARKNET_PATH}/examples/yolo.c
-    ${DARKNET_PATH}/examples/go.c
+    ${DARKNET_PATH}/src/activation_layer.c        ${DARKNET_PATH}/src/activations.c
+    ${DARKNET_PATH}/src/art.c                     ${DARKNET_PATH}/src/avgpool_layer.c
+    ${DARKNET_PATH}/src/batchnorm_layer.c         ${DARKNET_PATH}/src/blas.c
+    ${DARKNET_PATH}/src/box.c                     ${DARKNET_PATH}/src/captcha.c
+    ${DARKNET_PATH}/src/cifar.c                   ${DARKNET_PATH}/src/classifier.c
+    ${DARKNET_PATH}/src/coco.c                    ${DARKNET_PATH}/src/col2im.c
+    ${DARKNET_PATH}/src/compare.c                 ${DARKNET_PATH}/src/connected_layer.c
+    ${DARKNET_PATH}/src/conv_lstm_layer.c         ${DARKNET_PATH}/src/convolutional_layer.c
+    ${DARKNET_PATH}/src/cost_layer.c              ${DARKNET_PATH}/src/cpu_gemm.c
+    ${DARKNET_PATH}/src/crnn_layer.c              ${DARKNET_PATH}/src/crop_layer.c
+    ${DARKNET_PATH}/src/dark_cuda.c               ${DARKNET_PATH}/src/data.c  
+    ${DARKNET_PATH}/src/deconvolutional_layer.c   ${DARKNET_PATH}/src/detection_layer.c
+    ${DARKNET_PATH}/src/demo.c                    
+    ${DARKNET_PATH}/src/detector.c                ${DARKNET_PATH}/src/dice.c
+    ${DARKNET_PATH}/src/dropout_layer.c           ${DARKNET_PATH}/src/gaussian_yolo_layer.c
+    ${DARKNET_PATH}/src/gemm.c                    ${DARKNET_PATH}/src/getopt.c
+    ${DARKNET_PATH}/src/gettimeofday.c            ${DARKNET_PATH}/src/go.c
+    ${DARKNET_PATH}/src/gru_layer.c               ${DARKNET_PATH}/src/http_stream.cpp
+    ${DARKNET_PATH}/src/im2col.c                  ${DARKNET_PATH}/src/image.c
+    ${DARKNET_PATH}/src/image_opencv.cpp          ${DARKNET_PATH}/src/layer.c
+    ${DARKNET_PATH}/src/list.c                    ${DARKNET_PATH}/src/local_layer.c
+    ${DARKNET_PATH}/src/lstm_layer.c              ${DARKNET_PATH}/src/matrix.c
+    ${DARKNET_PATH}/src/maxpool_layer.c           ${DARKNET_PATH}/src/network.c
+    ${DARKNET_PATH}/src/nightmare.c               ${DARKNET_PATH}/src/normalization_layer.c
+    ${DARKNET_PATH}/src/option_list.c             ${DARKNET_PATH}/src/parser.c
+    ${DARKNET_PATH}/src/region_layer.c            ${DARKNET_PATH}/src/reorg_layer.c
+    ${DARKNET_PATH}/src/reorg_old_layer.c         ${DARKNET_PATH}/src/rnn.c
+    ${DARKNET_PATH}/src/rnn_layer.c               ${DARKNET_PATH}/src/rnn_vid.c
+    ${DARKNET_PATH}/src/route_layer.c             ${DARKNET_PATH}/src/representation_layer.c
+    ${DARKNET_PATH}/src/sam_layer.c               ${DARKNET_PATH}/src/scale_channels_layer.c
+    ${DARKNET_PATH}/src/shortcut_layer.c          ${DARKNET_PATH}/src/softmax_layer.c           
+    ${DARKNET_PATH}/src/super.c                   ${DARKNET_PATH}/src/swag.c
+    ${DARKNET_PATH}/src/tag.c                     ${DARKNET_PATH}/src/tree.c
+    ${DARKNET_PATH}/src/upsample_layer.c          ${DARKNET_PATH}/src/utils.c
+    ${DARKNET_PATH}/src/voxel.c                   ${DARKNET_PATH}/src/writing.c                 
+    ${DARKNET_PATH}/src/yolo.c                    ${DARKNET_PATH}/src/yolo_layer.c              
 )
 
 set(DARKNET_CUDA_FILES
-    ${DARKNET_PATH}/src/activation_kernels.cu     ${DARKNET_PATH}/src/crop_layer_kernels.cu
-    ${DARKNET_PATH}/src/avgpool_layer_kernels.cu  ${DARKNET_PATH}/src/deconvolutional_kernels.cu
-    ${DARKNET_PATH}/src/blas_kernels.cu           ${DARKNET_PATH}/src/dropout_layer_kernels.cu
-    ${DARKNET_PATH}/src/col2im_kernels.cu         ${DARKNET_PATH}/src/im2col_kernels.cu
-    ${DARKNET_PATH}/src/convolutional_kernels.cu  ${DARKNET_PATH}/src/maxpool_layer_kernels.cu
+    ${DARKNET_PATH}/src/activation_kernels.cu       ${DARKNET_PATH}/src/avgpool_layer_kernels.cu
+    ${DARKNET_PATH}/src/blas_kernels.cu             ${DARKNET_PATH}/src/col2im_kernels.cu
+    ${DARKNET_PATH}/src/convolutional_kernels.cu    ${DARKNET_PATH}/src/crop_layer_kernels.cu
+    ${DARKNET_PATH}/src/deconvolutional_kernels.cu  ${DARKNET_PATH}/src/dropout_layer_kernels.cu
+    ${DARKNET_PATH}/src/im2col_kernels.cu           ${DARKNET_PATH}/src/maxpool_layer_kernels.cu
+    ${DARKNET_PATH}/src/network_kernels.cu
 )
 
-if (CUDA_FOUND)
+if (CUDA_SUCCESS)
 
   link_directories(
     ${CUDA_TOOLKIT_ROOT_DIR}/lib64
@@ -157,6 +226,14 @@ if (CUDA_FOUND)
     curand
   )
 
+  if(CUDNN_FOUND)
+    target_link_libraries(${PROJECT_NAME}_lib CuDNN::CuDNN)
+    target_compile_definitions(${PROJECT_NAME}_lib PUBLIC -DCUDNN)
+    if(ENABLE_CUDNN_HALF)
+      target_compile_definitions(${PROJECT_NAME}_lib PUBLIC -DCUDNN_HALF)
+    endif()
+  endif()
+
   cuda_add_executable(${PROJECT_NAME}
     src/yolo_object_detector_node.cpp
   )
@@ -240,6 +317,15 @@ if (NOT EXISTS "${FILE}")
   execute_process(COMMAND wget -q https://github.com/leggedrobotics/darknet_ros/releases/download/1.1.4/yolov3.weights -P ${PATH})
 endif()
 
+#Download yolov4.weights
+
+set(FILE "${PATH}/yolov4.weights")
+message(STATUS "Checking and downloading yolov4.weights if needed ...")
+if (NOT EXISTS "${FILE}")
+  message(STATUS "... file does not exist. Downloading now ...")
+  execute_process(COMMAND wget -q https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v3_optimal/yolov4.weights -P ${PATH})
+endif()
+
 #############
 ## Testing ##
 #############
@@ -280,3 +366,4 @@ if (cmake_clang_tools_FOUND)
     CF_WERROR
   )
 endif (cmake_clang_tools_FOUND)
+

darknet_ros/config/ros.yaml

@@ -1,7 +1,7 @@
 subscribers:
 
   camera_reading:
-    topic: /camera/rgb/image_raw
+    topic: /cv_camera/image_raw
     queue_size: 1
 
 actions:

darknet_ros/config/yolov4-tiny.yaml

@@ -0,0 +1,90 @@
+yolo_model:
+
+  config_file:
+    name: yolov4-tiny.cfg
+  weight_file:
+    name: yolov4-tiny.weights
+  threshold:
+    value: 0.3
+  detection_classes:
+    names:
+      - person
+      - bicycle
+      - car
+      - motorbike
+      - aeroplane
+      - bus
+      - train
+      - truck
+      - boat
+      - traffic light
+      - fire hydrant
+      - stop sign
+      - parking meter
+      - bench
+      - bird
+      - cat
+      - dog
+      - horse
+      - sheep
+      - cow
+      - elephant
+      - bear
+      - zebra
+      - giraffe
+      - backpack
+      - umbrella
+      - handbag
+      - tie
+      - suitcase
+      - frisbee
+      - skis
+      - snowboard
+      - sports ball
+      - kite
+      - baseball bat
+      - baseball glove
+      - skateboard
+      - surfboard
+      - tennis racket
+      - bottle
+      - wine glass
+      - cup
+      - fork
+      - knife
+      - spoon
+      - bowl
+      - banana
+      - apple
+      - sandwich
+      - orange
+      - broccoli
+      - carrot
+      - hot dog
+      - pizza
+      - donut
+      - cake
+      - chair
+      - sofa
+      - pottedplant
+      - bed
+      - diningtable
+      - toilet
+      - tvmonitor
+      - laptop
+      - mouse
+      - remote
+      - keyboard
+      - cell phone
+      - microwave
+      - oven
+      - toaster
+      - sink
+      - refrigerator
+      - book
+      - clock
+      - vase
+      - scissors
+      - teddy bear
+      - hair drier
+      - toothbrush