feat: add yolov10 node

perception/autoware_tensorrt_yolov10/CMakeLists.txt

@@ -0,0 +1,177 @@
+cmake_minimum_required(VERSION 3.17)
+project(autoware_tensorrt_yolov10)
+
+find_package(autoware_tensorrt_common)
+if(NOT ${autoware_tensorrt_common_FOUND})
+  message(WARNING "The autoware_tensorrt_common package is not found. Please check its dependencies.")
+  return()
+endif()
+
+find_package(autoware_cmake REQUIRED)
+autoware_package()
+
+add_compile_options(-Wno-deprecated-declarations)
+
+find_package(OpenCV REQUIRED)
+
+option(CUDA_VERBOSE "Verbose output of CUDA modules" OFF)
+
+# set flags for CUDA availability
+option(CUDA_AVAIL "CUDA available" OFF)
+find_package(CUDA)
+if(CUDA_FOUND)
+  find_library(CUBLAS_LIBRARIES cublas HINTS
+    ${CUDA_TOOLKIT_ROOT_DIR}/lib64
+    ${CUDA_TOOLKIT_ROOT_DIR}/lib
+  )
+  if(CUDA_VERBOSE)
+    message("CUDA is available!")
+    message("CUDA Libs: ${CUDA_LIBRARIES}")
+    message("CUDA Headers: ${CUDA_INCLUDE_DIRS}")
+  endif()
+  # Note: cublas_device was depreciated in CUDA version 9.2
+  #       https://forums.developer.nvidia.com/t/where-can-i-find-libcublas-device-so-or-libcublas-device-a/67251/4
+  #       In LibTorch, CUDA_cublas_device_LIBRARY is used.
+  unset(CUDA_cublas_device_LIBRARY CACHE)
+  set(CUDA_AVAIL ON)
+else()
+  message("CUDA NOT FOUND")
+  set(CUDA_AVAIL OFF)
+endif()
+
+# set flags for TensorRT availability
+option(TRT_AVAIL "TensorRT available" OFF)
+# try to find the tensorRT modules
+find_library(NVINFER nvinfer)
+find_library(NVONNXPARSER nvonnxparser)
+if(NVINFER AND NVONNXPARSER)
+  if(CUDA_VERBOSE)
+    message("TensorRT is available!")
+    message("NVINFER: ${NVINFER}")
+    message("NVONNXPARSER: ${NVONNXPARSER}")
+  endif()
+  set(TRT_AVAIL ON)
+else()
+  message("TensorRT is NOT Available")
+  set(TRT_AVAIL OFF)
+endif()
+
+# set flags for CUDNN availability
+option(CUDNN_AVAIL "CUDNN available" OFF)
+# try to find the CUDNN module
+find_library(CUDNN_LIBRARY
+NAMES libcudnn.so${__cudnn_ver_suffix} libcudnn${__cudnn_ver_suffix}.dylib ${__cudnn_lib_win_name}
+PATHS $ENV{LD_LIBRARY_PATH} ${__libpath_cudart} ${CUDNN_ROOT_DIR} ${PC_CUDNN_LIBRARY_DIRS} ${CMAKE_INSTALL_PREFIX}
+PATH_SUFFIXES lib lib64 bin
+DOC "CUDNN library."
+)
+if(CUDNN_LIBRARY)
+  if(CUDA_VERBOSE)
+    message(STATUS "CUDNN is available!")
+    message(STATUS "CUDNN_LIBRARY: ${CUDNN_LIBRARY}")
+  endif()
+  set(CUDNN_AVAIL ON)
+else()
+  message("CUDNN is NOT Available")
+  set(CUDNN_AVAIL OFF)
+endif()
+
+find_package(OpenMP)
+if(OpenMP_FOUND)
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
+endif()
+
+##########
+# tensorrt_yolov10
+ament_auto_add_library(${PROJECT_NAME} SHARED
+  src/tensorrt_yolov10.cpp
+  src/tensorrt_yolov10_node.cpp
+)
+
+ament_target_dependencies(${PROJECT_NAME}
+  OpenCV
+)
+
+if(TRT_AVAIL AND CUDA_AVAIL AND CUDNN_AVAIL)
+  # Officially, add_library supports .cu file compilation.
+  # However, as of cmake 3.22.1, it seems to fail compilation because compiler flags for
+  # C++ are directly passed to nvcc (they are originally space separated
+  # but nvcc assume comma separated as argument of `-Xcompiler` option).
+  # That is why `cuda_add_library` is used here.
+#   cuda_add_library(${PROJECT_NAME}_gpu_preprocess
+#     SHARED
+#     src/preprocess.cu
+#   )
+
+#   target_include_directories(${PROJECT_NAME}_gpu_preprocess PUBLIC
+#     $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
+#     $<INSTALL_INTERFACE:include/${PROJECT_NAME}>
+#   )
+
+#   target_link_libraries(${PROJECT_NAME}
+#     ${autoware_tensorrt_common_LIBRARIES}
+#     ${PROJECT_NAME}_gpu_preprocess
+#   )
+else()
+  target_link_libraries(${PROJECT_NAME}
+    ${autoware_tensorrt_common_LIBRARIES}
+  )
+endif()
+
+target_compile_definitions(${PROJECT_NAME} PRIVATE
+  TENSORRT_VERSION_MAJOR=${TENSORRT_VERSION_MAJOR}
+)
+
+ament_auto_add_library(yolov10_single_image_inference_node SHARED
+  src/yolov10_single_image_inference_node.cpp
+)
+
+ament_target_dependencies(yolov10_single_image_inference_node
+  OpenCV
+)
+
+target_link_libraries(yolov10_single_image_inference_node
+  ${PROJECT_NAME}
+  stdc++fs
+)
+
+target_compile_definitions(yolov10_single_image_inference_node PRIVATE
+  TENSORRT_VERSION_MAJOR=${TENSORRT_VERSION_MAJOR}
+)
+
+rclcpp_components_register_node(yolov10_single_image_inference_node
+  PLUGIN "autoware::tensorrt_yolov10::Yolov10SingleImageInferenceNode"
+  EXECUTABLE yolov10_single_image_inference
+)
+
+ament_auto_add_library(${PROJECT_NAME}_node SHARED
+  src/tensorrt_yolov10_node.cpp
+)
+
+ament_target_dependencies(${PROJECT_NAME}_node
+  OpenCV
+)
+
+target_link_libraries(${PROJECT_NAME}_node
+  ${PROJECT_NAME}
+)
+
+target_compile_definitions(${PROJECT_NAME}_node PRIVATE
+  TENSORRT_VERSION_MAJOR=${TENSORRT_VERSION_MAJOR}
+)
+
+rclcpp_components_register_node(${PROJECT_NAME}_node
+  PLUGIN "autoware::tensorrt_yolov10::TrtYolov10Node"
+  EXECUTABLE ${PROJECT_NAME}_node_exe
+)
+
+if(BUILD_TESTING)
+  find_package(ament_lint_auto REQUIRED)
+  ament_lint_auto_find_test_dependencies()
+endif()
+
+ament_auto_package(INSTALL_TO_SHARE
+  launch
+  config
+)

perception/autoware_tensorrt_yolov10/README.md

@@ -0,0 +1,136 @@
+# autoware_tensorrt_yolov10
+
+## Purpose
+
+This package detects target objects e.g., cars, trucks, bicycles, pedestrians,etc on a image based on [YOLOV10](https://github.com/THU-MIG/yolov10) model.
+
+## Inputs / Outputs
+
+### Input
+
+| Name       | Type                | Description     |
+| ---------- | ------------------- | --------------- |
+| `in/image` | `sensor_msgs/Image` | The input image |
+
+### Output
+
+| Name          | Type                                               | Description                                        |
+| ------------- | -------------------------------------------------- | -------------------------------------------------- |
+| `out/objects` | `tier4_perception_msgs/DetectedObjectsWithFeature` | The detected objects with 2D bounding boxes        |
+| `out/image`   | `sensor_msgs/Image`                                | The image with 2D bounding boxes for visualization |
+
+## Assumptions / Known limits
+
+The label contained in detected 2D bounding boxes (i.e., `out/objects`) will be either one of the followings:
+
+- CAR
+- PEDESTRIAN ("PERSON" will also be categorized as "PEDESTRIAN")
+- BUS
+- TRUCK
+- BICYCLE
+- MOTORCYCLE
+
+If other labels (case insensitive) are contained in the file specified via the `label_file` parameter,
+those are labeled as `UNKNOWN`, while detected rectangles are drawn in the visualization result (`out/image`).
+
+## Onnx model
+
+you can download yolov10m.onnx from [releases](https://github.com/THU-MIG/yolov10/releases)
+
+## Label file
+
+This file represents the correspondence between class index (integer outputted from YOLOV10 network) and
+class label (strings making understanding easier). This package maps class IDs (incremented from 0)
+with labels according to the order in this file.
+
+currently, this file is actually a coco label which contains the following labels:
+
+```text
+person
+bicycle
+car
+motorcycle
+airplane
+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
+couch
+potted plant
+bed
+dining table
+toilet
+tv
+laptop
+mouse
+remote
+keyboard
+cell phone
+microwave
+oven
+toaster
+sink
+refrigerator
+book
+clock
+vase
+scissors
+teddy bear
+hair drier
+```
+
+## Reference repositories
+
+- <https://github.com/THU-MIG/yolov10>
+
+## Legal Notice
+
+The inference code is licensed under Apache 2.0. The model and training code are licensed under AGPL-3.0. you can check details from <https://github.com/THU-MIG/yolov10?tab=AGPL-3.0-1-ov-file>. To inquire about a commercial license when using trained model weights please contact yolov10 author.

perception/autoware_tensorrt_yolov10/config/yolov10.param.yaml

@@ -0,0 +1,41 @@
+# cspell:ignore semseg
+/**:
+  ros__parameters:
+
+    # refine segmentation mask by overlay roi class
+    # disable when sematic segmentation accuracy is good enough
+    is_roi_overlap_segment: true
+
+    # minimum existence_probability of detected roi considered to replace segmentation
+    overlap_roi_score_threshold: 0.3
+
+    # publish color mask for result visualization
+    is_publish_color_mask: false
+
+    roi_overlay_segment_label:
+      UNKNOWN : true
+      CAR : false
+      TRUCK : false
+      BUS : false
+      MOTORCYCLE : true
+      BICYCLE : true
+      PEDESTRIAN : true
+      ANIMAL: true
+
+    image_path: "$(var data_path)/tensorrt_yolov10/000000000307.jpg"
+    model_path: "$(var data_path)/tensorrt_yolov10/$(var model_name).onnx" # The onnx file name for YOLOX model.
+    label_path: "$(var data_path)/tensorrt_yolov10/label.txt" # The label file path for YOLOX model.
+    color_map_path: "$(var data_path)/tensorrt_yolov10/semseg_color_map.csv"
+    score_threshold: 0.35 # Objects with a score lower than this value will be ignored. This threshold will be ignored if specified model contains EfficientNMS_TRT module in it.
+    nms_threshold: 0.7 # Detection results will be ignored if IoU over this value. This threshold will be ignored if specified model contains EfficientNMS_TRT module in it.
+
+    precision: "fp16" # Operation precision to be used on inference. Valid value is one of: [fp32, fp16, int8].
+    calibration_algorithm: "MinMax" # Calibration algorithm to be used for quantization when precision==int8. Valid value is one of: [Entropy, (Legacy | Percentile), MinMax].
+    dla_core_id: -1 # If positive ID value is specified, the node assign inference task to the DLA core.
+    quantize_first_layer: false # If true, set the operating precision for the first (input) layer to be fp16. This option is valid only when precision==int8.
+    quantize_last_layer: false # If true, set the operating precision for the last (output) layer to be fp16. This option is valid only when precision==int8.
+    profile_per_layer: false # If true, profiler function will be enabled. Since the profile function may affect execution speed, it is recommended to set this flag true only for development purpose.
+    clip_value: 0.0 # If positive value is specified, the value of each layer output will be clipped between [0.0, clip_value]. This option is valid only when precision==int8 and used to manually specify the dynamic range instead of using any calibration.
+    preprocess_on_gpu: true # If true, pre-processing is performed on GPU.
+    gpu_id: 0 # GPU ID to select CUDA Device
+    calibration_image_list_path: "" # Path to a file which contains path to images. Those images will be used for int8 quantization.