initialied the repository

KalanaRatnayake · Jul 10, 2024 · 83e1491 · 83e1491
commit 83e1491
Show file tree

Hide file tree

Showing 20 changed files with 913 additions and 0 deletions.
diff --git a/.github/workflows/boxmot_humble_amd64_arm64.yml b/.github/workflows/boxmot_humble_amd64_arm64.yml
@@ -0,0 +1,61 @@
+name: ROS Humble BOXMOT AMD64/ARM64
+
+on:
+  workflow_dispatch:
+
+  push:
+    branches: 
+      - main
+    paths-ignore:
+      - README.md
+
+  pull_request:
+    branches: 
+      - main
+    paths-ignore:
+      - README.md
+
+env:
+  REGISTRY: ghcr.io
+  OWNER: kalanaratnayake
+  IMAGE_NAME: boxmot-ros
+
+# https://docs.github.com/en/actions/using-jobs/using-concurrency
+concurrency:
+  # only cancel in-progress jobs or runs for the current workflow - matches against branch & tags
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+
+  build-and-push-image:
+    runs-on: ubuntu-latest
+
+    permissions:
+      contents: write
+      packages: write
+
+    steps:      
+      - name: Check out the repository
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to the Container registry
+        run: |
+          docker login --username ${{ env.OWNER }} --password ${{ secrets.GH_PAT }} ghcr.io 
+
+      - name: Build and Push
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          file: docker/Dockerfile
+          platforms: linux/amd64
+          push: true
+          tags: |
+            ${{ env.REGISTRY }}/${{ env.OWNER }}/${{ env.IMAGE_NAME }}:latest
+            ${{ env.REGISTRY }}/${{ env.OWNER }}/${{ env.IMAGE_NAME }}:humble
diff --git a/.github/workflows/boxmot_humble_j_nano.yml b/.github/workflows/boxmot_humble_j_nano.yml
@@ -0,0 +1,61 @@
+name: ROS Humble BOXMOT Jetson Nano
+
+on:
+  workflow_dispatch:
+
+  push:
+    branches: 
+      - main
+    paths-ignore:
+      - README.md
+
+  pull_request:
+    branches: 
+      - main
+    paths-ignore:
+      - README.md
+
+env:
+  REGISTRY: ghcr.io
+  OWNER: kalanaratnayake
+  IMAGE_NAME: boxmot-ros
+
+# https://docs.github.com/en/actions/using-jobs/using-concurrency
+concurrency:
+  # only cancel in-progress jobs or runs for the current workflow - matches against branch & tags
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+
+  build-and-push-image:
+    runs-on: ubuntu-latest
+
+    permissions:
+      contents: write
+      packages: write
+
+    steps:
+      - name: Check out the repository
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to the Container registry
+        run: |
+          docker login --username ${{ env.OWNER }} --password ${{ secrets.GH_PAT }} ghcr.io 
+
+      - name: Build and Push
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          file: docker/Dockerfile.jetson-nano
+          platforms: linux/arm64
+          push: true
+          tags: |
+            ${{ env.REGISTRY }}/${{ env.OWNER }}/${{ env.IMAGE_NAME }}:latest-j-nano
+            ${{ env.REGISTRY }}/${{ env.OWNER }}/${{ env.IMAGE_NAME }}:humble-j-nano
diff --git a/README.md b/README.md
@@ -0,0 +1,179 @@
+# boxmot_ros
+
+## Docker Usage by adding to compose.yml file
+
+To use GPU with docker while on AMD64 systems, install [nvidia-container-toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html) with given instructions.
+
+### Supported platforms
+
+Replace `image` and 'device' parameter in the compose.yml with following values for respective systems.
+
+| System              | ROS Version | Value for `image`                                 | Value for `device`  | Size    |
+| :---                | :---        | :---                                              |  :---               | :---:   |
+| AMD64               | Humble      | ghcr.io/kalanaratnayake/boxmot-ros:humble         | `cpu`, `0`, `0,1,2` | 5.64 GB |
+| Jetson Nano         | Humble      | ghcr.io/kalanaratnayake/boxmot-ros:humble-j-nano  | `cpu`, `0`          | 3.29GB  |
+
+### Default models with Docker Compose
+
+Add the following snippet under `services` to any compose.yaml file to add this container and use an existing model.
+
+```bash
+services:
+  boxmot:
+    image: ghcr.io/kalanaratnayake/boxmot-ros:humble
+    environment:
+      - TRACKING_MODEL=weights/yolov9t.pt
+      - REID_MODEL=weights/yolov9t.pt
+      - INPUT_TOPIC=/yolo_ros/detection_result
+      - PUBLISH_ANNOTATED_IMAGE=True
+      - OUTPUT_ANNOTATED_TOPIC=/boxmot_ros/annotated_image
+      - OUTPUT_DETAILED_TOPIC=/boxmot_ros/tracking_result
+      - CONFIDENCE_THRESHOLD=0.25
+      - DEVICE='0'
+    restart: unless-stopped
+    privileged: true
+    network_mode: host
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]   
+    volumes:
+      - boxmot:/boxmot
+
+volumes:
+  boxmot:
+```
+
+### Custom models with Docker Compose
+
+Add the following snippet under `services` to any compose.yaml file to add this container and use an existing model.
+
+```bash
+services:
+  boxmot:
+    image: ghcr.io/kalanaratnayake/boxmot-ros:humble
+    environment:
+      - TRACKING_MODEL=weights/yolov9t.pt
+      - REID_MODEL=weights/yolov9t.pt
+      - INPUT_TOPIC=/yolo_ros/detection_result
+      - PUBLISH_ANNOTATED_IMAGE=True
+      - OUTPUT_ANNOTATED_TOPIC=/boxmot_ros/annotated_image
+      - OUTPUT_DETAILED_TOPIC=/boxmot_ros/tracking_result
+      - CONFIDENCE_THRESHOLD=0.25
+      - DEVICE='0'
+    restart: unless-stopped
+    privileged: true
+    network_mode: host
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]   
+    volumes:
+      - type: bind
+        source: /home/kalana/Downloads/weights
+        target: /boxmot/weights
+
+volumes:
+  boxmot:
+```
+
+## Docker Usage with this repository
+
+Clone this reposiotory
+
+```bash
+mkdir -p boxmot_ws/src && cd boxmot_ws/src
+git clone https://github.com/KalanaRatnayake/boxmot_ros.git && cd ..
+```
+
+<details> 
+<summary> <h3> on AMD64 </h3> </summary>
+
+Pull the Docker image and start compose (No need to run `docker compose build`)
+```bash
+cd src/boxmot_ros/docker
+docker compose -f compose.amd64.yaml pull
+docker compose -f compose.amd64.yaml up
+```
+</details>
+
+<details> 
+<summary> <h3> on JetsonNano </h3> </summary>
+
+Pull the Docker image and start compose (No need to run `docker compose build`)
+```bash
+cd src/boxmot_ros/docker
+docker compose -f compose.jnano.yaml pull
+docker compose -f compose.jnano.yaml up
+```
+</details>
+
+<br>
+
+## Native Usage
+
+Clone this repository with and install dependencies.
+
+```bash
+git clone https://github.com/KalanaRatnayake/boxmot_ros.git
+git clone https://github.com/KalanaRatnayake/boxmot_ros_msgs.git
+cd boxmot_ros
+pip3 install -r requirements.txt
+```
+
+### Build the package
+
+If required, edit the parameters at `config/boxmot_ros_params.yaml' and then at the workspace root run,
+```bash
+colcon build
+```
+### Start the system
+
+To use the launch file, run,
+
+```bash
+source ./install/setup.bash
+ros2 launch boxmot_ros boxmot.launch.py
+```
+
+<br>
+<br>
+
+## Parameter decription
+
+| ROS Parameter           | Docker ENV parameter    | Default Value                | Description |
+| :---                    | :---                    | :---:                        | :---        |
+| tracking_model          | TRACKING_MODEL          | `deepocsort`                 | Model to be used for tracking. see [1] for default models and [2] for custom models |
+| reid_model              | REID_MODEL              | `yolov9t.pt`                 | Model to be used for reidentification. see [1] for default models and [2] for custom models |
+| input_topic             | INPUT_TOPIC             | `/yolo_ros/detection_result` | Topic to subscribe for RGB image. Accepts `sensor_msgs/Image` |
+| publish_annotated_image | PUBLISH_ANNOTATED_IMAGE | `False`                      | Whether to publish annotated image, increases callback execution time when set to `True` |
+| output_annotated_topic  | OUTPUT_ANNOTATED_TOPIC  | `/boxmot_ros/annotated_image` | Topic for publishing annotated images uses `sensor_msgs/Image` |
+| output_detailed_topic   | OUTPUT_DETAILED_TOPIC   | `/boxmot_ros/tracking_result` | Topic for publishing detailed results uses `boxmot_ros_msgs/YoloResult` |
+| confidence_threshold    | CONFIDENCE_THRESHOLD    | `0.25`                      | Confidence threshold for predictions |
+| device                  | DEVICE                  | `'0'`                       | `cpu` for CPU, `0` for gpu, `0,1,2,3` if there are multiple GPUs |
+
+
+[1] If the reid model is available at [MODEL_ZOO](https://kaiyangzhou.github.io/deep-person-reid/MODEL_ZOO), and tracking_model is supported [deepocsort, strongsort, ocsort, bytetrack, botsort]. They will be downloaded from the cloud at the startup. We are using docker volumes to maintain downloaded weights so that weights are not downloaded at each startup. Use the snipped in [Default models with Docker Compose](https://github.com/KalanaRatnayake/boxmot_ros#default-models-with-docker-compose)
+
+[2] Give the tracking model weight file's name as `TRACKING_MODEL` parameter and reidentification model weight file's name as `REID_MODEL` parameter. Update the docker volume source tag to direct to the folder and use docker bind-mounts instead of docker volumes where the weight files exist in the host machine. As an example if the weight files are in `/home/kalana/Downloads/weight/` folder, then use the snipped in [Custom models with Docker Compose](https://github.com/KalanaRatnayake/boxmot_ros#custom-models-with-docker-compose)
+
+## Latency description
+
+Here is a summary of whether supported models work with boxmot_ros node (in docker) on various platforms and the time it takes to execute a single interation of `YoloROS.image_callback` function. Values are measured as an average of 100 executions of the function and Input is a 640x480 RGB image at 30 fps.
+
+Laptop -> Ryzen 9 16 core with RTX3070 mobile GPU with Ubuntu 22.04
+Jetson Nano -> Overclocked with [Qengineering Ubuntu 20.04 in Headless mode](https://github.com/Qengineering/Jetson-Nano-Ubuntu-20-image?tab=readme-ov-file#headless)
+
+| Model | Laptop (amd64) | Jetson Nano |
+| :---  |  ---: | ---: |
+| `deepocsort` | 27 ms |  975 ms |
+| `strongsort` | 20 ms |  800 ms |
+| `ocsort`     | 17 ms |  750 ms |
+| `bytetrack`  | 17 ms |  650 ms |
+| `botsort`    | 14 ms |  210 ms |
diff --git a/boxmot_ros/__init__.py b/boxmot_ros/__init__.py