By clicking “Accept All Cookies”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage and assist in our marketing efforts. More info
Cookie Settings
By clicking “Accept All Cookies”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage and assist in our marketing efforts. More info
Learn how to export Ultralytics YOLO models like Ultralytics YOLO11 with PaddlePaddle for efficient deployment across edge, mobile, and cloud platforms.
As artificial intelligence (AI) advances, machines are getting better at understanding the world around them. One key area driving this progress is computer vision, a branch of AI that enables machines to interpret and make decisions based on visual data.
From helping cars recognize traffic signs to checking shelves in retail stores, computer vision is now part of many everyday tools. These tasks rely on Vision AI models that can quickly scan a photo or video and identify what’s important.
Over time, these models have become faster and more accurate, making them useful in areas like farming, healthcare, security, and retail. For example, Ultralytics YOLO11 is a model built to handle a range of computer vision tasks with speed and precision. It can detect and classify objects, track movement, and estimate body poses.
A vital part of moving computer vision from research into real-world applications is deployment. Once a model has been trained, the next step is to run it on devices such as phones, edge hardware, or cloud servers.
Fig 1. Model deployment is a key part of any computer vision project.
To support this, Ultralytics YOLO models like YOLO11 can be exported into various formats depending on the target platform. One of these formats is PaddlePaddle, an open-source AI framework that enables efficient model deployment and inference across a wide range of devices and systems.
In this article, we’ll explore how Ultralytics YOLO11 can be exported through the PaddlePaddle integration supported by Ultralytics to enable efficient deployment across various platforms.
What is PaddlePaddle?
Deploying AI models outside of research settings, like on mobile devices or edge hardware, can be tricky at times, especially when you need them to run efficiently and use minimal resources. PaddlePaddle is a deep learning framework designed to help with exactly that.
It is a Chinese open-source platform, with its name standing for Parallel Distributed Deep Learning. Developed by Baidu, a company well known for its work in AI and software infrastructure, PaddlePaddle was created especially for real-world applications, not just research.
Developers can run models in PaddlePaddle format on servers, edge devices, and even mobile hardware. It also supports tools that simplify AI development, including low-code and no-code options. The platform has a strong developer community of over 4.7 million users, and is used across a variety of industries, including healthcare, agriculture, manufacturing, and finance.
Key features of PaddlePaddle
Here are some of the key features that help PaddlePaddle run models more efficiently on real-world devices:
Dynamic-to-static graph conversion: This feature turns a flexible model into a fixed version that runs more smoothly and predictably. A fixed model is easier to optimize and faster when making predictions.
Operator fusion: PaddlePaddle can combine multiple steps in the model into one. This reduces how much memory the model uses and helps it run faster. Think of it like combining several tasks into a single action to save time.
Quantization: This makes the model smaller by using simpler numbers (like rounding to fewer decimal places). It helps the model run on devices with limited power, like phones or smart cameras, without losing much accuracy.
Fig 2. Advantages of using PaddlePaddle. Image by author.
An overview of YOLO11 deployment with PaddlePaddle
The PaddlePaddle integration supported by Ultralytics makes it easier to move from training to deployment. Developers who already use PaddlePaddle tools can bring YOLO11 into their workflows more easily.
The Ultralytics Python package supports direct export of YOLO11 models to the PaddlePaddle format, allowing developers to deploy trained models without extra tools or manual conversion steps.
The export process can be done using either the command line or Python code, so developers can choose whichever method fits best with their workflow. This helps keep things simple and reduces the chance of setup issues. Once exported, the model can be used for computer vision tasks like object detection, image classification, pose estimation and instance segmentation.
It is great option for deployment scenarios where devices have limited memory or require fast processing. The exported models are optimized to run efficiently, even on resource-constrained systems.
How to export YOLO11 models to PaddlePaddle format
It only takes a few steps to export YOLO11 to PaddlePaddle model format.
The first step is to install the Ultralytics Python package using a package manager like ‘pip’. This can be done by running the command “pip install ultralytics” in your command prompt or terminal to get started.
The Ultralytics package provides tools for training, evaluating, fine-tuning, exporting, and deploying models for a range of computer vision tasks. If you run into any issues during installation, check the Common Issues guide for troubleshooting tips.
Once your environment is set up, you can load and export a pre-trained YOLO11 model such as “yolo11n.pt” as shown below. You can also export your own custom-trained YOLO11 model.
After your model has been converted to PaddlePaddle format, it can be deployed in various scenarios on different types of hardware.
For instance, in the example below, we are loading a YOLO11 model that was exported to PaddlePaddle format and using it to make a prediction. This process, known as running an inference, simply means using the model to analyze new data. Here, we're testing it with an image of two dogs.
After running the code, the output image with the model’s predictions will be saved automatically in the “runs/detect/predict” folder.
Fig 3. Using the exported YOLO11 model to detect objects in an image. Image by Author.
Deploying YOLO11 using the PaddlePaddle framework
PaddlePaddle offers several deployment tools, each suited for different devices and use cases like cloud environments, embedded systems, and web applications. Here are some of the main deployment options:
Paddle Serving: It helps deploy models as REST APIs, making it a good choice for cloud or server environments that need features like version control and online testing.
Paddle Inference API: It gives you more control over how models run, which is helpful when you need to fine-tune performance or build custom application logic.
Paddle Lite: It is designed for lightweight deployment on mobile devices, tablets, and embedded systems. It’s optimized for smaller models and faster inference on hardware with limited resources.
Paddle.js: It lets you run AI models in web browsers using technologies like WebGL and WebAssembly, which is useful for interactive demos and browser-based tools.
Fig 4. Deployment options enabled by PaddlePaddle. Image by author.
Once you pick the right tool for your setup, you can load the exported model. The PaddlePaddle engine takes care of the next steps. It loads the model, processes the input image, and gives back the results.
When should you choose the PaddlePaddle integration?
The Ultralytics Python package also supports various other export formats, so you may find yourself asking: When is PaddlePaddle the right choice?
PaddlePaddle is a reliable option when you're looking to deploy models on devices with limited resources, such as smartphones, embedded systems, or edge hardware. It's also great for real-time applications that need fast, efficient performance, like object detection in mobile apps, vision-based monitoring in smart cameras, or pose estimation running directly on-device without cloud support.
Beyond this, if the project needs to run offline or in low-connectivity environments, you can consider using the PaddlePaddle integration. Applications like visual inspection tools in manufacturing, handheld devices for field surveys, or AI-enabled retail scanners can benefit from PaddlePaddle’s lightweight runtime and flexible deployment options.
PaddlePaddle limitations to consider
While PaddlePaddle offers interesting deployment capabilities, here are some constraining factors to be aware of:
Smaller global community: Outside of China, the user and contributor base is relatively small. This can make it harder to find community support, resolved GitHub issues, or Stack Overflow answers.
Steeper learning curve for non-Baidu tools: PaddlePaddle integrates smoothly with Baidu’s ecosystem, but using it outside that context may involve extra configuration and setup steps.
Fewer integrations with mainstream ML tools: PaddlePaddle has limited compatibility with common tools like Hugging Face Transformers, MLflow, or Kubernetes-native AI services.
Key takeaways
The PaddlePaddle integration supported by Ultralytics makes it easy to export and deploy YOLO11 models across a range of devices. It's especially useful for projects that require efficient, on-device performance - like mobile apps, smart cameras, or embedded systems.
With just a few steps, you can bring powerful vision models into real-world applications. As computer vision continues to advance, tools like YOLO and PaddlePaddle are making it easier than ever to build fast, intelligent systems across everything from consumer devices to industrial tools.
EN
ZH
KO
JA
RU
DE
FR
ES
PT
IT
TR
VI
AR
