ultralytics 8.0.195 NVIDIA Triton Inference Server support (#5257)

Co-authored-by: TheConstant3 <46416203+TheConstant3@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

docs/guides/azureml-quickstart.md

@@ -77,7 +77,7 @@ Train a detection model for 10 epochs with an initial learning_rate of 0.01:
 yolo train data=coco128.yaml model=yolov8n.pt epochs=10 lr0=0.01
 ```
 
-You can find more [instructions to use the Ultralytics cli here](https://docs.ultralytics.com/quickstart/#use-ultralytics-with-cli).
+You can find more [instructions to use the Ultralytics CLI here](https://docs.ultralytics.com/quickstart/#use-ultralytics-with-cli).
 
 ## Quickstart from a Notebook
 

docs/guides/index.md

@@ -22,6 +22,7 @@ Here's a compilation of in-depth guides to help you master different aspects of
 * [Conda Quickstart](conda-quickstart.md) 🚀 NEW: Step-by-step guide to setting up a [Conda](https://anaconda.org/conda-forge/ultralytics) environment for Ultralytics. Learn how to install and start using the Ultralytics package efficiently with Conda.
 * [Docker Quickstart](docker-quickstart.md) 🚀 NEW: Complete guide to setting up and using Ultralytics YOLO models with [Docker](https://hub.docker.com/r/ultralytics/ultralytics). Learn how to install Docker, manage GPU support, and run YOLO models in isolated containers for consistent development and deployment.
 * [Raspberry Pi](raspberry-pi.md) 🚀 NEW: Quickstart tutorial to run YOLO models to the latest Raspberry Pi hardware.
+* [Triton Inference Server Integration](triton-inference-server.md) 🚀 NEW: Dive into the integration of Ultralytics YOLOv8 with NVIDIA's Triton Inference Server for scalable and efficient deep learning inference deployments.
 
 ## Contribute to Our Guides
 

docs/guides/raspberry-pi.md

@@ -37,47 +37,25 @@ You should see a video feed from your camera.
 
 This guide offers you the flexibility to start with either [YOLOv5](https://github.com/ultralytics/yolov5) or [YOLOv8](https://github.com/ultralytics/ultralytics). Both versions have their unique advantages and use-cases. The choice is yours, but remember, the guide's aim is not just quick setup but also a robust foundation for your future work in object detection.
 
-## Hardware Specifics: Raspberry Pi 3 vs Raspberry Pi 4
+## Hardware Specifics: At a Glance
 
-Raspberry Pi 3 and Raspberry Pi 4 have distinct hardware specifications, and the YOLO installation and configuration process can vary slightly depending on which model you're using.
+To assist you in making an informed hardware decision, we've summarized the key hardware specifics of Raspberry Pi 3, 4, and 5 in the table below:
 
-### Raspberry Pi 3
-
-- **CPU**: 1.2GHz Quad-Core ARM Cortex-A53
-- **RAM**: 1GB LPDDR2
-- **USB Ports**: 4 x USB 2.0
-- **Network**: Ethernet & Wi-Fi 802.11n
-- **Performance**: Generally slower, may require lighter YOLO models for real-time processing
-- **Power Requirement**: 2.5A power supply
-- **Official Documentation**: [Raspberry Pi 3 Documentation](https://www.raspberrypi.org/documentation/hardware/raspberrypi/bcm2837/README.md)
-
-### Raspberry Pi 4
-
-- **CPU**: 1.5GHz Quad-core 64-bit ARM Cortex-A72 CPU
-- **RAM**: Options of 2GB, 4GB or 8GB LPDDR4
-- **USB Ports**: 2 x USB 2.0, 2 x USB 3.0
-- **Network**: Gigabit Ethernet & Wi-Fi 802.11ac
-- **Performance**: Faster, capable of running more complex YOLO models in real-time
-- **Power Requirement**: 3.0A USB-C power supply
-- **Official Documentation**: [Raspberry Pi 4 Documentation](https://www.raspberrypi.org/documentation/hardware/raspberrypi/bcm2711/README.md)
-
-### Raspberry Pi 5
-
-- **CPU**: 2.4GHz Quad-core 64-bit Arm Cortex-A76 CPU
-- **GPU**: VideoCore VII, supporting OpenGL ES 3.1, Vulkan 1.2
-- **Display Output**: Dual 4Kp60 HDMI
-- **Decoder**: 4Kp60 HEVC
-- **Network**: Gigabit Ethernet with PoE+ support, Dual-band 802.11ac Wi-Fi®, Bluetooth 5.0 / BLE
-- **USB Ports**: 2 x USB 3.0, 2 x USB 2.0
-- **Other Features**: High-speed microSD card interface with SDR104 mode, 2 × 4-lane MIPI camera/display transceivers, PCIe 2.0 x1 interface, standard 40-pin GPIO header, real-time clock, power button
-- **Power Requirement**: Specifics not yet available, expected to require a higher amperage supply
-- **Official Documentation**: [Raspberry Pi 5 Documentation](https://www.raspberrypi.com/news/introducing-raspberry-pi-5/)
+| Feature                    | Raspberry Pi 3                                                                           | Raspberry Pi 4                                                                           | Raspberry Pi 5                                                       |
+|----------------------------|------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------|----------------------------------------------------------------------|
+| **CPU**                    | 1.2GHz Quad-Core ARM Cortex-A53                                                          | 1.5GHz Quad-core 64-bit ARM Cortex-A72                                                   | 2.4GHz Quad-core 64-bit Arm Cortex-A76                               |
+| **RAM**                    | 1GB LPDDR2                                                                               | 2GB, 4GB or 8GB LPDDR4                                                                   | *Details not yet available*                                          |
+| **USB Ports**              | 4 x USB 2.0                                                                              | 2 x USB 2.0, 2 x USB 3.0                                                                 | 2 x USB 3.0, 2 x USB 2.0                                             |
+| **Network**                | Ethernet & Wi-Fi 802.11n                                                                 | Gigabit Ethernet & Wi-Fi 802.11ac                                                        | Gigabit Ethernet with PoE+ support, Dual-band 802.11ac Wi-Fi®        |
+| **Performance**            | Slower, may require lighter YOLO models                                                  | Faster, can run complex YOLO models                                                      | *Details not yet available*                                          |
+| **Power Requirement**      | 2.5A power supply                                                                        | 3.0A USB-C power supply                                                                  | *Details not yet available*                                          |
+| **Official Documentation** | [Link](https://www.raspberrypi.org/documentation/hardware/raspberrypi/bcm2837/README.md) | [Link](https://www.raspberrypi.org/documentation/hardware/raspberrypi/bcm2711/README.md) | [Link](https://www.raspberrypi.com/news/introducing-raspberry-pi-5/) |
 
 Please make sure to follow the instructions specific to your Raspberry Pi model to ensure a smooth setup process.
 
 ## Quick Start with YOLOv5
 
-This section outlines how to set up YOLOv5 on a Raspberry Pi 3 or 4 with a Pi Camera. These steps are designed to be compatible with the libcamera camera stack introduced in Raspberry Pi OS Bullseye.
+This section outlines how to set up YOLOv5 on a Raspberry Pi with a Pi Camera. These steps are designed to be compatible with the libcamera camera stack introduced in Raspberry Pi OS Bullseye.
 
 ### Install Necessary Packages
 
@@ -171,7 +149,7 @@ Follow this section if you are interested in setting up YOLOv8 instead. The step
     sudo apt-get autoremove -y
     ```
 
-2. Install YOLOv8:
+2. Install the `ultralytics` Python package:
 
     ```bash
     pip3 install ultralytics
@@ -183,28 +161,6 @@ Follow this section if you are interested in setting up YOLOv8 instead. The step
     sudo reboot
     ```
 
-### Modify `build.py`
-
-Just like YOLOv5, YOLOv8 also needs minor modifications to accept TCP streams.
-
-1. Open `build.py` located in the Ultralytics package folder:
-
-    ```bash
-    sudo nano /home/pi/.local/lib/pythonX.X/site-packages/ultralytics/build.py
-    ```
-
-2. Find and modify the `is_url` line to accept TCP streams:
-
-    ```python
-    is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://', 'tcp://'))
-    ```
-
-3. Save and exit:
-
-    ```bash
-    CTRL + O -> ENTER -> CTRL + X
-    ```
-
 ### Initiate TCP Stream with Libcamera
 
 1. Start the TCP stream:
@@ -231,7 +187,7 @@ while True:
 
 ## Next Steps
 
-Congratulations on successfully setting up YOLO on your Raspberry Pi! For further learning and support, visit [Ultralytics](https://ultralytics.com/) and [KashmirWorldFoundation](https://www.kashmirworldfoundation.org/).
+Congratulations on successfully setting up YOLO on your Raspberry Pi! For further learning and support, visit [Ultralytics](https://ultralytics.com/) and [Kashmir World Foundation](https://www.kashmirworldfoundation.org/).
 
 ## Acknowledgements and Citations
 

docs/guides/triton-inference-server.md

@@ -0,0 +1,137 @@
+---
+comments: true
+description: A step-by-step guide on integrating Ultralytics YOLOv8 with Triton Inference Server for scalable and high-performance deep learning inference deployments.
+keywords: YOLOv8, Triton Inference Server, ONNX, Deep Learning Deployment, Scalable Inference, Ultralytics, NVIDIA, Object Detection, Cloud Inferencing
+---
+
+# Triton Inference Server with Ultralytics YOLOv8
+
+The [Triton Inference Server](https://developer.nvidia.com/nvidia-triton-inference-server) (formerly known as TensorRT Inference Server) is an open-source software solution developed by NVIDIA. It provides a cloud inferencing solution optimized for NVIDIA GPUs. Triton simplifies the deployment of AI models at scale in production. Integrating Ultralytics YOLOv8 with Triton Inference Server allows you to deploy scalable, high-performance deep learning inference workloads. This guide provides steps to set up and test the integration.
+
+<p align="center">
+  <br>
+  <iframe width="720" height="405" src="https://www.youtube.com/embed/NQDtfSi5QF4" 
+    title="Getting Started with NVIDIA Triton Inference Server" frameborder="0" 
+    allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" 
+    allowfullscreen>
+  </iframe>
+  <br>
+  <strong>Watch:</strong> Getting Started with NVIDIA Triton Inference Server.
+</p>
+
+## What is Triton Inference Server?
+
+Triton Inference Server is designed to deploy a variety of AI models in production. It supports a wide range of deep learning and machine learning frameworks, including TensorFlow, PyTorch, ONNX Runtime, and many others. Its primary use cases are:
+
+- Serving multiple models from a single server instance.
+- Dynamic model loading and unloading without server restart.
+- Ensemble inferencing, allowing multiple models to be used together to achieve results.
+- Model versioning for A/B testing and rolling updates.
+
+## Prerequisites
+
+Ensure you have the following prerequisites before proceeding:
+
+- Docker installed on your machine.
+- Install `tritonclient`:
+  ```bash
+  pip install tritonclient[all]
+  ```
+
+## Exporting YOLOv8 to ONNX Format
+
+Before deploying the model on Triton, it must be exported to the ONNX format. ONNX (Open Neural Network Exchange) is a format that allows models to be transferred between different deep learning frameworks. Use the `export` function from the `YOLO` class:
+
+```python
+from ultralytics import YOLO
+
+# Load a model
+model = YOLO('yolov8n.pt')  # load an official model
+
+# Export the model
+onnx_file = model.export(format='onnx', dynamic=True)
+```
+
+## Setting Up Triton Model Repository
+
+The Triton Model Repository is a storage location where Triton can access and load models.
+
+1. Create the necessary directory structure:
+
+    ```python
+    from pathlib import Path
+    
+    # Define paths
+    triton_repo_path = Path('tmp') / 'triton_repo'
+    triton_model_path = triton_repo_path / 'yolo'
+    
+    # Create directories
+    (triton_model_path / '1').mkdir(parents=True, exist_ok=True)
+    ```
+
+2. Move the exported ONNX model to the Triton repository:
+
+    ```python
+    from pathlib import Path
+    
+    # Move ONNX model to Triton Model path
+    Path(onnx_file).rename(triton_model_path / '1' / 'model.onnx')
+    
+    # Create config file
+    (triton_model_path / 'config.pdtxt').touch()
+    ```
+
+## Running Triton Inference Server
+
+Run the Triton Inference Server using Docker:
+
+```python
+import subprocess
+import time
+
+from tritonclient.http import InferenceServerClient
+
+# Define image https://catalog.ngc.nvidia.com/orgs/nvidia/containers/tritonserver
+tag = 'nvcr.io/nvidia/tritonserver:23.09-py3'  # 6.4 GB
+
+# Pull the image
+subprocess.call(f'docker pull {tag}', shell=True)
+
+# Run the Triton server and capture the container ID
+container_id = subprocess.check_output(
+    f'docker run -d --rm -v {triton_repo_path}:/models -p 8000:8000 {tag} tritonserver --model-repository=/models',
+    shell=True).decode('utf-8').strip()
+
+# Wait for the Triton server to start
+triton_client = InferenceServerClient(url='localhost:8000', verbose=False, ssl=False)
+
+# Wait until model is ready
+for _ in range(10):
+    with contextlib.suppress(Exception):
+        assert triton_client.is_model_ready(model_name)
+        break
+    time.sleep(1)
+```
+
+Then run inference using the Triton Server model:
+
+```python
+from ultralytics import YOLO
+
+# Load the Triton Server model
+model = YOLO(f'http://localhost:8000/yolo', task='detect')
+
+# Run inference on the server
+results = model('path/to/image.jpg')
+```
+
+Cleanup the container:
+
+```python
+# Kill and remove the container at the end of the test
+subprocess.call(f'docker kill {container_id}', shell=True)
+```
+
+---
+
+By following the above steps, you can deploy and run Ultralytics YOLOv8 models efficiently on Triton Inference Server, providing a scalable and high-performance solution for deep learning inference tasks. If you face any issues or have further queries, refer to the [official Triton documentation](https://docs.nvidia.com/deeplearning/triton-inference-server/user-guide/docs/index.html) or reach out to the Ultralytics community for support.

docs/modes/export.md

@@ -57,7 +57,7 @@ Export a YOLOv8n model to a different format like ONNX or TensorRT. See Argument
 
         # Load a model
         model = YOLO('yolov8n.pt')  # load an official model
-        model = YOLO('path/to/best.pt')  # load a custom trained
+        model = YOLO('path/to/best.pt')  # load a custom trained model
 
         # Export the model
         model.export(format='onnx')

docs/reference/utils/triton.md

@@ -0,0 +1,9 @@
+# Reference for `ultralytics/utils/triton.py`
+
+!!! note
+
+    Full source code for this file is available at [https://github.com/ultralytics/ultralytics/blob/main/ultralytics/utils/triton.py](https://github.com/ultralytics/ultralytics/blob/main/ultralytics/utils/triton.py). Help us fix any issues you see by submitting a [Pull Request](https://docs.ultralytics.com/help/contributing/) 🛠️. Thank you 🙏!
+
+---
+## ::: ultralytics.utils.triton.TritonRemoteModel
+<br><br>

docs/tasks/classify.md

@@ -140,7 +140,7 @@ Export a YOLOv8n-cls model to a different format like ONNX, CoreML, etc.
 
         # Load a model
         model = YOLO('yolov8n-cls.pt')  # load an official model
-        model = YOLO('path/to/best.pt')  # load a custom trained
+        model = YOLO('path/to/best.pt')  # load a custom trained model
 
         # Export the model
         model.export(format='onnx')

docs/tasks/detect.md

@@ -152,7 +152,7 @@ Export a YOLOv8n model to a different format like ONNX, CoreML, etc.
 
         # Load a model
         model = YOLO('yolov8n.pt')  # load an official model
-        model = YOLO('path/to/best.pt')  # load a custom trained
+        model = YOLO('path/to/best.pt')  # load a custom trained model
 
         # Export the model
         model.export(format='onnx')

docs/tasks/pose.md

@@ -156,7 +156,7 @@ Export a YOLOv8n Pose model to a different format like ONNX, CoreML, etc.
 
         # Load a model
         model = YOLO('yolov8n-pose.pt')  # load an official model
-        model = YOLO('path/to/best.pt')  # load a custom trained
+        model = YOLO('path/to/best.pt')  # load a custom trained model
 
         # Export the model
         model.export(format='onnx')

docs/tasks/segment.md

@@ -157,7 +157,7 @@ Export a YOLOv8n-seg model to a different format like ONNX, CoreML, etc.
 
         # Load a model
         model = YOLO('yolov8n-seg.pt')  # load an official model
-        model = YOLO('path/to/best.pt')  # load a custom trained
+        model = YOLO('path/to/best.pt')  # load a custom trained model
 
         # Export the model
         model.export(format='onnx')