pranaypalem/yolov8_tensorrt repository overview

⁠YOLOv8 TensorRT ROS Inference

Minimal, high-performance Docker image for YOLOv8 object detection optimized with TensorRT and integrated into ROS Noetic for robotics workflows.

⁠🛠️ Key Features

• YOLOv8 + TensorRT optimized inference
• ROS Noetic integration (subscribes to image topics, publishes annotated detections)
• CUDA 11.8 + cuDNN8 GPU acceleration
• ONNX + TensorRT engine auto-conversion
• Real-time detection for robotic applications
• CPU fallback if GPU not available
• Preinstalled with Ultralytics, ONNX Runtime, and OpenCV

⁠📦 Usage

1. Run container:

docker run -it --gpus all --net=host \
  -e ROS_MASTER_URI=http://<ROS_MASTER_IP>:11311 \
  -e ROS_IP=<YOUR_LOCAL_IP> \
  pranaypalem/yolov8_tensorrt:3.1.3

1. Convert YOLO model to TensorRT engine inside container:

python3 convert_to_engine.py

1. Start ROS inference:

python3 yolo_ros_inference.py

⁠📋 Requirements

• ROS Noetic running (remote or same machine)
• TensorRT tar.gz file (download manually from NVIDIA TensorRT⁠)
• YOLOv8 .pt model (example: bestsynthetic.pt)

⁠📂 Default Workspace

/workspace
├── bestsynthetic.pt
├── convert_to_engine.py
├── yolo_ros_inference.py
├── coke_can_depth_processor.py
├── video_publisher.py
├── test_video.avi
└── TensorRT tar.gz

⁠🧑‍💻 Author

Pranay Palem – Focused on real-time robotics, computer vision pipelines, and deployable AI systems.

⁠📜 License

MIT License — Free for commercial and personal use.