Vehicle Detection with YOLOv8

🏁 Introduction

YOLOv8 is a real-time object detection model developed by Ultralytics. This repository demonstrate how to train YOLOv8 on KITTI dataset and use it to detect vehicles in images and videos. Then we will deploy the trained model as an API server using FastAPI.

👀 Quickstart

Quickstart will guide you to run vehicle detection API server using Docker. If you want to train YOLOv8 on KITTI dataset, please refer to the Installation section.

We assume that you have Docker installed. First, we need to download the trained model from Release page. You can download the model by running the following command:

cd tmp
wget https://github.com/ruhyadi/vehicle-detection-yolov8/releases/download/v0.0/vehicle_kitti_v0_last.pt
cd ..

You should have the model in the tmp folder.

Next, we need to build the Docker image. To build the Docker image, run the following command:

docker build -f dockerfile.api -t ruhyadi/vehicle-detection-api:latest .

You should have the Docker image with the name ruhyadi/vehicle-detection-api:latest.

Now, we can run the Docker container. To run the Docker container, run the following command:

bash scripts/start_api.sh

# output
vehicle-yolov8-api  | Available providers: ['CPUExecutionProvider']
vehicle-yolov8-api  | Setup YOLOv8 ONNX engine...
vehicle-yolov8-api  | Setup YOLOv8 ONNX engine... Done
vehicle-yolov8-api  | Starting uvicorn server on 0.0.0.0:7000...
vehicle-yolov8-api  | INFO:     Started server process [1]
vehicle-yolov8-api  | INFO:     Waiting for application startup.
vehicle-yolov8-api  | INFO:     Application startup complete.
vehicle-yolov8-api  | INFO:     Uvicorn running on http://0.0.0.0:7000 (Press CTRL+C to quit)

You can access the API server Swagger at http://localhost:7000.

📥 Installation

📦 Create a virtual environment

We assume that you have Anaconda installed. To install the required packages, run the following commands:

conda create -n yolov8 python=3.10 cudatoolkit=11.8
conda activate yolov8

🐍 Install PyTorch

Next, install PyTorch with the following command:

# CUDA 11.8
pip install torch==2.0.0+cu118 torchvision==0.15.1+cu118 --index-url https://download.pytorch.org/whl/cu118

# CPU only
pip install torch==2.0.0+cpu torchvision==0.15.1+cpu --index-url https://download.pytorch.org/whl/cpu

🚓 Download KITTI dataset

Download the KITTI dataset and extract it to the data folder.

You should have the following folder structure:

data
├── kitti
│   ├── image_2 # images from the left color camera
│   │   ├── 000000.png
│   │   ├── 000001.png
│   │   ├── ...
│   │   └── 007480.png
│   ├── label_2 # label files in KITTI format
│   │   ├── 000000.txt
│   │   ├── 000001.txt
│   │   ├── ...
│   │   └── 007480.txt
│   ├── calib # calibration files
│   │   ├── 000000.txt
|   │   ├── 000001.txt
│   │   ├── ...
│   │   └── 007480.txt

In order to train YOLOv8 with KITTI dataset, the first step we need to rename image_2 to images. You can do this by running the following command:

mv data/kitti/image_2 data/kitti/images

You should have the following folder structure:

data
├── kitti
│   ├── images # images from the left color camera
│   │   ├── 000000.png
│   │   ├── 000001.png
│   │   ├── ...
│   │   └── 007480.png
│   ├── label_2 # label files in KITTI format
│   │   ├── 000000.txt
│   │   ├── 000001.txt
│   │   ├── ...
│   │   └── 007480.txt
│   ├── calib # calibration files
│   │   ├── 000000.txt
|   │   ├── 000001.txt
│   │   ├── ...
│   │   └── 007480.txt

🔧 Convert KITTI format to YOLO format

KITTI dataset uses a different format than YOLO. To convert the KITTI format to YOLO format, run the following command:

python tools/kitti2yolo.py \
    --images_dir data/kitti/images \
    --labels_dir data/kitti/label_2 \
    --output_dir data/kitti

You should have the following folder structure:

data
├── kitti
│   ├── images
|   ├── label_2
│   ├── calib
│   └── labels # label files in YOLO format
│       ├── 000000.txt
│       ├── 000001.txt
│       ├── ...
│       └── 007480.txt

🪛 Create YOLO Training and Validation sets

To create the YOLO training and validation sets, run the following command:

python scripts/generate_yolo_sets.py \
    --images_dir data/kitti/images \
    --output_dir data/kitti \
    --train_val_split 0.80 \
    --prefix yolo

You should have the following folder structure:

data
├── kitti
│   ├── images
|   ├── label_2
│   ├── calib
│   ├── labels
│   ├── yolo_train.txt # YOLO training set
│   └── yolo_val.txt # YOLO validation set

🖨️ Create a YAML Configuration File

Create a YAML configuration file for training. You can use the dataset/yolov8s.yml file as a template. The configuration file should be placed in the configs folder.

🥋 Training

All sets are ready. Now, we can start training. To train YOLOv8, run the following command:

python src/train.py \
    --weights yolov8s.pt \
    --config configs/dataset/kitti.yaml \
    --epochs 15 \
    --batch-size 4 \
    --img-size 640 \
    --device 0 \
    --workers 4

If you want to resume training from a checkpoint, add the --resume flag:

python src/train.py \
    --weights yolov8s.pt \
    --config configs/dataset/kitti.yaml \
    --epochs 15 \
    --batch-size 4 \
    --img-size 640 \
    --device 0 \
    --workers 4 \
    --resume

You can also use the --weights flag to specify a custom weight file.

🔋 Export to ONNX

ONNX (Open Neural Network Exchange) is an open format for representing deep learning models. To export the trained model to ONNX, run the following command:

python tools/torch2onnx.py \
    --weights_path tmp/vehicle_kitti_v0_last.pt

You should have exported ONNX model in the tmp folder.

💌 Akwnowledgement

• Ultralytics YOLOv8: YOLOv8 is a real-time object detection model developed by Ultralytics.
• FastAPI: FastAPI is a modern, fast (high-performance), web framework for building APIs with Python 3.6+ based on standard Python type hints.