Merge pull request #57 from Interplai/add_yolov3

Adds yolov3 model using tvm to AutowareAuto

Author: LucaFos

perception/camera_obstacle_detection/yolo_v3/tensorflow_fp32_coco/README.md

@@ -0,0 +1,88 @@
+# YOLO V3 Darknet Conversion to Keras
+
+The network definition and weights come from [darknet
+website](https://pjreddie.com/darknet/yolo/). It is converted to the keras model using [this](https://github.com/qqwweee/keras-yolo3) repository.
+
+It has been converted to onnx format
+using [tf2onnx](https://github.com/onnx/tensorflow-onnx).
+
+## Executing the model 
+
+All commands should be run from the root of the model zoo directory.
+
+1.Compile a local image of `autoware/model-zoo-tvm-cli`
+
+```bash
+$./scripts/tvm_cli/build.sh
+```
+
+2.Compile the model by running the TVM CLI script in a docker container
+
+```bash
+$ MODEL_DIR=$(pwd)/perception/camera_obstacle_detection/yolo_v3/tensorflow_fp32_coco/
+$ export MODEL_DIR
+$ docker run \
+    -it --rm \
+    -v ${MODEL_DIR}:${MODEL_DIR} -w ${MODEL_DIR} \
+    -u $(id -u ${USER}):$(id -g ${USER}) \
+    autoware/model-zoo-tvm-cli:local \
+        compile \
+        --config ${MODEL_DIR}/definition.yaml \
+        --output_path ${MODEL_DIR}/execute_model/build
+```
+
+3.Compile the c++ pipeline
+
+```bash
+$ docker run \
+    -it --rm \
+    -v ${MODEL_DIR}:${MODEL_DIR} -w ${MODEL_DIR}/execute_model/build \
+    -u $(id -u ${USER}):$(id -g ${USER}) \
+    --entrypoint "" \
+    autoware/model-zoo-tvm-cli:local \
+    bash -c "cmake .. && make -j"
+```
+
+4.Download a sample image and copy some files needed for decoding detections
+
+```bash
+$ curl https://raw.githubusercontent.com/pjreddie/darknet/master/data/dog.jpg \
+    > ${MODEL_DIR}/execute_model/build/test_image_0.jpg
+$ cp ${MODEL_DIR}/model_files/labels.txt ${MODEL_DIR}/execute_model/build/
+$ cp ${MODEL_DIR}/model_files/anchors.csv ${MODEL_DIR}/execute_model/build/
+```
+
+5.Run the detection pipeline inside a docker container. The output result can be obtained in two ways:
+
+- **Save as an image**: saves the result of the pipeline as an image file in the build directory, the filename `output.jpg` can be changed in the command if needed:
+
+    ```bash
+    $ docker run \
+        -it --rm \
+        --net=host \
+        -v ${MODEL_DIR}:${MODEL_DIR} \
+        -w ${MODEL_DIR}/execute_model/build \
+        --entrypoint "" \
+        autoware/model-zoo-tvm-cli:local \
+            ./execute_model output.jpg
+    ```
+
+- **Display in a X11 window**: X draw calls are forwarded to the host so the detection results can be displayed in a X11 window.
+
+    ```bash
+    $ docker run \
+        -it --rm \
+        -v /tmp/.X11-unix:/tmp/.X11-unix:rw \
+        -v ${HOME}/.Xauthority:${HOME}/.Xauthority:rw \
+        -e XAUTHORITY=${HOME}/.Xauthority \
+        -e DISPLAY=$DISPLAY \
+        --net=host \
+        -v ${MODEL_DIR}:${MODEL_DIR} \
+        -w ${MODEL_DIR}/execute_model/build \
+        --entrypoint "" \
+        autoware/model-zoo-tvm-cli:local \
+            ./execute_model
+    ```
+
+For more information about getting the TVM docker image, see the TVM CLI
+[documentation](../../../../scripts/tvm_cli/README.md).

perception/camera_obstacle_detection/yolo_v3/tensorflow_fp32_coco/definition.yaml

@@ -0,0 +1,43 @@
+version: 1
+enable_testing: true
+network:
+  filename: ./model_files/yolov3-416.onnx
+  framework: ONNX
+  provenance: ./README.md
+  training: COCO dataset, https://pjreddie.com/darknet/yolo/
+  model_license: Apache-2.0
+  data_license: CC-BY-4.0
+network_parameters:
+  datatype: float32
+  input_nodes:
+    - name: input
+      description: Camera Image RGB
+      shape:
+        - 1
+        - 416
+        - 416
+        - 3
+  output_nodes:
+    - name: conv2d_58
+      description:
+      shape:
+        - 1
+        - 13
+        - 13
+        - 255
+    
+    - name: conv2d_66
+      description:
+      shape:
+        - 1
+        - 26
+        - 26
+        - 255
+
+    - name: conv2d_74
+      description:
+      shape:
+        - 1
+        - 52
+        - 52
+        - 255

perception/camera_obstacle_detection/yolo_v3/tensorflow_fp32_coco/execute_model/CMakeLists.txt

@@ -0,0 +1,47 @@
+# Copyright 2021 Apex.AI, Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+cmake_minimum_required(VERSION 3.0)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+set(CMAKE_CXX_STANDARD 14)
+
+project(execute_model)
+
+find_package( OpenCV REQUIRED )
+
+file(GLOB SRC_FILES "*.cpp")
+
+set(TVM_ROOT /usr/tvm)
+set(DMLC_CORE ${TVM_ROOT}/3rdparty/dmlc-core)
+set(DLPACK ${TVM_ROOT}/3rdparty/dlpack)
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -O3")
+
+add_executable(${CMAKE_PROJECT_NAME}
+    ${ROS_NODE_FILE}
+    ${SRC_FILES}
+)
+
+target_link_libraries(${CMAKE_PROJECT_NAME}
+    ${OpenCV_LIBS}
+    dl
+    pthread
+    tvm_runtime
+)
+
+target_include_directories (${CMAKE_PROJECT_NAME} PRIVATE
+    ${TVM_ROOT}/include
+    ${DMLC_CORE}/include
+    ${DLPACK}/include
+)