Deploy YOLOv11 with NVIDIA DeepStream 7.1

This repository enables real-time object detection using YOLOv11 integrated into the NVIDIA DeepStream SDK (7.1) via the official container from NGC (NVIDIA GPU Cloud).

---

Project Structure

deepstream_yolov11/
├── Dockerfile                            # Docker container setup based on DeepStream 7.1
├── setup.sh                              # Script to export YOLOv11 to ONNX and generate TRT engine
├── deepstream_app_config.txt             # DeepStream application pipeline config
├── config_infer_primary_yoloV11.txt      # YOLOv11 model configuration for DeepStream inference
├── weights/
│   └── yolo11s.pt                        # 🔁 Place your YOLOv11 PyTorch model weights here manually
└── README.md                             # This file

---

Objective

• Convert YOLOv11 PyTorch model to ONNX
• Integrate into DeepStream 7.1 pipeline
• Run object detection on video using optimized TensorRT engine

---

Requirements

• NVIDIA GPU with drivers installed
• Docker
• NVIDIA Container Toolkit
• Linux system (tested on Ubuntu 22.04)
• Model weights: yolo11s.pt from Ultralytics

---

Setup Instructions

1. Clone This Repo

git clone https://github.com/T-DevH/deepstream_yolov11.git
cd deepstream_yolov11

---

2. Place Model Weights

Manually download yolo11s.pt from the official Ultralytics release page and place it in the weights/ folder:

mv ~/Downloads/yolo11s.pt weights/

---

3. Build the Docker Container

docker build -t deepstream-yolov11 .

---

4. Run the Container

Allows GUI video display by forwarding X11 and mounting video source

xhost +local:root

docker run --gpus all -it --rm   --net=host   --runtime=nvidia   -e DISPLAY=$DISPLAY   -v /tmp/.X11-unix:/tmp/.X11-unix   -v $(pwd):/workspace   deepstream-yolov11

---

5. Inside the Container

Run the setup script to:

• Export yolo11s.pt → ONNX
• Build a TensorRT .engine
• Prepare DeepStream inference pipeline

./setup.sh

---

6. Run DeepStream Application

deepstream-app -c deepstream_app_config.txt

You should see real-time object detection running on the provided video (edit deepstream_app_config.txt to change video source).

---

How DeepStream-Yolo Is Used

This project integrates with the excellent DeepStream-Yolo project to:

• ✅ Export YOLOv11 models from PyTorch to ONNX using utils/export_yoloV8.py
• ✅ Compile a custom DeepStream plugin libnvdsinfer_custom_impl_Yolo.so to parse YOLO outputs
• ✅ Reuse tried-and-tested configuration formats for DeepStream compatibility

These steps are handled automatically:

• The repository is cloned in the Dockerfile.
• The plugin is compiled during container build.
• The ONNX export is performed in setup.sh.
• The .so plugin is referenced inside config_infer_primary_yoloV11.txt.

No manual steps are needed — everything is automated within the Docker container.

---

License

This project uses open-source components under their respective licenses (YOLO, DeepStream, Ultralytics).

---

Acknowledgements

• Ultralytics
• NVIDIA DeepStream SDK
• MarcosLuciano’s DeepStream-Yolo

---

Need Help?

Feel free to open an issue or reach out if you need help with DeepStream, YOLO export, or pipeline tuning!