Intel OpenVINO Export

In this guide, we cover exporting YOLOv8 models to the OpenVINO format, which can provide up to 3x CPU speedup, as well as accelerating YOLO inference on Intel GPU and NPU hardware.

OpenVINO, short for Open Visual Inference & Neural Network Optimization toolkit, is a comprehensive toolkit for optimizing and deploying AI inference models. Even though the name contains Visual, OpenVINO also supports various additional tasks including language, audio, time series, etc.

Watch: How To Export and Optimize an Ultralytics YOLOv8 Model for Inference with OpenVINO.

Usage Examples

Export a YOLOv8n model to OpenVINO format and run inference with the exported model.

from ultralytics import YOLO

# Load a YOLOv8n PyTorch model
model = YOLO("yolov8n.pt")

# Export the model
model.export(format="openvino")  # creates 'yolov8n_openvino_model/'

# Load the exported OpenVINO model
ov_model = YOLO("yolov8n_openvino_model/")

# Run inference
results = ov_model("https://ultralytics.com/images/bus.jpg")

# Export a YOLOv8n PyTorch model to OpenVINO format
yolo export model=yolov8n.pt format=openvino  # creates 'yolov8n_openvino_model/'

# Run inference with the exported model
yolo predict model=yolov8n_openvino_model source='https://ultralytics.com/images/bus.jpg'

Arguments

Benefits of OpenVINO

1. Performance: OpenVINO delivers high-performance inference by utilizing the power of Intel CPUs, integrated and discrete GPUs, and FPGAs.
2. Support for Heterogeneous Execution: OpenVINO provides an API to write once and deploy on any supported Intel hardware (CPU, GPU, FPGA, VPU, etc.).
3. Model Optimizer: OpenVINO provides a Model Optimizer that imports, converts, and optimizes models from popular deep learning frameworks such as PyTorch, TensorFlow, TensorFlow Lite, Keras, ONNX, PaddlePaddle, and Caffe.
4. Ease of Use: The toolkit comes with more than 80 tutorial notebooks (including YOLOv8 optimization) teaching different aspects of the toolkit.

OpenVINO Export Structure

When you export a model to OpenVINO format, it results in a directory containing the following:

1. XML file: Describes the network topology.
2. BIN file: Contains the weights and biases binary data.
3. Mapping file: Holds mapping of original model output tensors to OpenVINO tensor names.

You can use these files to run inference with the OpenVINO Inference Engine.

Using OpenVINO Export in Deployment

Once you have the OpenVINO files, you can use the OpenVINO Runtime to run the model. The Runtime provides a unified API to inference across all supported Intel hardware. It also provides advanced capabilities like load balancing across Intel hardware and asynchronous execution. For more information on running the inference, refer to the Inference with OpenVINO Runtime Guide.

Remember, you'll need the XML and BIN files as well as any application-specific settings like input size, scale factor for normalization, etc., to correctly set up and use the model with the Runtime.

In your deployment application, you would typically do the following steps:

1. Initialize OpenVINO by creating core = Core().
2. Load the model using the core.read_model() method.
3. Compile the model using the core.compile_model() function.
4. Prepare the input (image, text, audio, etc.).
5. Run inference using compiled_model(input_data).

For more detailed steps and code snippets, refer to the OpenVINO documentation or API tutorial.

OpenVINO YOLOv8 Benchmarks

YOLOv8 benchmarks below were run by the Ultralytics team on 4 different model formats measuring speed and accuracy: PyTorch, TorchScript, ONNX and OpenVINO. Benchmarks were run on Intel Flex and Arc GPUs, and on Intel Xeon CPUs at FP32 precision (with the half=False argument).

Intel Flex GPU

The Intel® Data Center GPU Flex Series is a versatile and robust solution designed for the intelligent visual cloud. This GPU supports a wide array of workloads including media streaming, cloud gaming, AI visual inference, and virtual desktop Infrastructure workloads. It stands out for its open architecture and built-in support for the AV1 encode, providing a standards-based software stack for high-performance, cross-architecture applications. The Flex Series GPU is optimized for density and quality, offering high reliability, availability, and scalability.

Benchmarks below run on Intel® Data Center GPU Flex 170 at FP32 precision.

This table represents the benchmark results for five different models (YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, YOLOv8x) across four different formats (PyTorch, TorchScript, ONNX, OpenVINO), giving us the status, size, mAP50-95(B) metric, and inference time for each combination.

Intel Arc GPU

Intel® Arc™ represents Intel's foray into the dedicated GPU market. The Arc™ series, designed to compete with leading GPU manufacturers like AMD and NVIDIA, caters to both the laptop and desktop markets. The series includes mobile versions for compact devices like laptops, and larger, more powerful versions for desktop computers.

The Arc™ series is divided into three categories: Arc™ 3, Arc™ 5, and Arc™ 7, with each number indicating the performance level. Each category includes several models, and the 'M' in the GPU model name signifies a mobile, integrated variant.

Early reviews have praised the Arc™ series, particularly the integrated A770M GPU, for its impressive graphics performance. The availability of the Arc™ series varies by region, and additional models are expected to be released soon. Intel® Arc™ GPUs offer high-performance solutions for a range of computing needs, from gaming to content creation.

Benchmarks below run on Intel® Arc 770 GPU at FP32 precision.

Intel Xeon CPU

The Intel® Xeon® CPU is a high-performance, server-grade processor designed for complex and demanding workloads. From high-end cloud computing and virtualization to artificial intelligence and machine learning applications, Xeon® CPUs provide the power, reliability, and flexibility required for today's data centers.

Notably, Xeon® CPUs deliver high compute density and scalability, making them ideal for both small businesses and large enterprises. By choosing Intel® Xeon® CPUs, organizations can confidently handle their most demanding computing tasks and foster innovation while maintaining cost-effectiveness and operational efficiency.

Benchmarks below run on 4th Gen Intel® Xeon® Scalable CPU at FP32 precision.

Intel Core CPU

The Intel® Core® series is a range of high-performance processors by Intel. The lineup includes Core i3 (entry-level), Core i5 (mid-range), Core i7 (high-end), and Core i9 (extreme performance). Each series caters to different computing needs and budgets, from everyday tasks to demanding professional workloads. With each new generation, improvements are made to performance, energy efficiency, and features.

Benchmarks below run on 13th Gen Intel® Core® i7-13700H CPU at FP32 precision.

Intel Ultra 7 155H Meteor Lake CPU

The Intel® Ultra™ 7 155H represents a new benchmark in high-performance computing, designed to cater to the most demanding users, from gamers to content creators. The Ultra™ 7 155H is not just a CPU; it integrates a powerful GPU and an advanced NPU (Neural Processing Unit) within a single chip, offering a comprehensive solution for diverse computing needs.

This hybrid architecture allows the Ultra™ 7 155H to excel in both traditional CPU tasks and GPU-accelerated workloads, while the NPU enhances AI-driven processes, enabling faster and more efficient machine learning operations. This makes the Ultra™ 7 155H a versatile choice for applications requiring high-performance graphics, complex computations, and AI inference.

The Ultra™ 7 series includes multiple models, each offering different levels of performance, with the 'H' designation indicating a high-power variant suitable for laptops and compact devices. Early benchmarks have highlighted the exceptional performance of the Ultra™ 7 155H, particularly in multitasking environments, where the combined power of the CPU, GPU, and NPU leads to remarkable efficiency and speed.

As part of Intel's commitment to cutting-edge technology, the Ultra™ 7 155H is designed to meet the needs of future computing, with more models expected to be released. The availability of the Ultra™ 7 155H varies by region, and it continues to receive praise for its integration of three powerful processing units in a single chip, setting new standards in computing performance.

Benchmarks below run on Intel® Ultra™ 7 155H at FP32 and INT8 precision.

Benchmarks

Integrated Intel® Arc™ GPUIntel® Meteor Lake CPUIntegrated Intel® AI Boost NPU

Model	Format	Precision	Status	Size (MB)	metrics/mAP50-95(B)	Inference time (ms/im)
YOLOv8n	PyTorch	FP32	✅	6.2	0.6381	35.95
YOLOv8n	OpenVINO	FP32	✅	12.3	0.6117	8.32
YOLOv8n	OpenVINO	INT8	✅	3.6	0.5791	9.88
YOLOv8s	PyTorch	FP32	✅	21.5	0.6967	79.72
YOLOv8s	OpenVINO	FP32	✅	42.9	0.7136	13.37
YOLOv8s	OpenVINO	INT8	✅	11.2	0.7086	9.96
YOLOv8m	PyTorch	FP32	✅	49.7	0.737	202.05
YOLOv8m	OpenVINO	FP32	✅	99.1	0.7331	28.07
YOLOv8m	OpenVINO	INT8	✅	25.5	0.7259	21.11
YOLOv8l	PyTorch	FP32	✅	83.7	0.7769	393.37
YOLOv8l	OpenVINO	FP32	✅	167.0	0.0	52.73
YOLOv8l	OpenVINO	INT8	✅	42.6	0.7861	28.11
YOLOv8x	PyTorch	FP32	✅	130.5	0.7759	610.71
YOLOv8x	OpenVINO	FP32	✅	260.6	0.748	73.51
YOLOv8x	OpenVINO	INT8	✅	66.0	0.8085	51.71

Model	Format	Precision	Status	Size (MB)	metrics/mAP50-95(B)	Inference time (ms/im)
YOLOv8n	PyTorch	FP32	✅	6.2	0.6381	34.69
YOLOv8n	OpenVINO	FP32	✅	12.3	0.6092	39.06
YOLOv8n	OpenVINO	INT8	✅	3.6	0.5968	18.37
YOLOv8s	PyTorch	FP32	✅	21.5	0.6967	79.9
YOLOv8s	OpenVINO	FP32	✅	42.9	0.7136	82.6
YOLOv8s	OpenVINO	INT8	✅	11.2	0.7083	29.51
YOLOv8m	PyTorch	FP32	✅	49.7	0.737	202.43
YOLOv8m	OpenVINO	FP32	✅	99.1	0.728	181.27
YOLOv8m	OpenVINO	INT8	✅	25.5	0.7285	51.25
YOLOv8l	PyTorch	FP32	✅	83.7	0.7769	385.87
YOLOv8l	OpenVINO	FP32	✅	167.0	0.7551	347.75
YOLOv8l	OpenVINO	INT8	✅	42.6	0.7675	91.66
YOLOv8x	PyTorch	FP32	✅	130.5	0.7759	603.63
YOLOv8x	OpenVINO	FP32	✅	260.6	0.7479	516.39
YOLOv8x	OpenVINO	INT8	✅	66.0	0.8119	142.42

Model	Format	Precision	Status	Size (MB)	metrics/mAP50-95(B)	Inference time (ms/im)
YOLOv8n	PyTorch	FP32	✅	6.2	0.6381	36.98
YOLOv8n	OpenVINO	FP32	✅	12.3	0.6103	16.68
YOLOv8n	OpenVINO	INT8	✅	3.6	0.5941	14.6
YOLOv8s	PyTorch	FP32	✅	21.5	0.6967	79.76
YOLOv8s	OpenVINO	FP32	✅	42.9	0.7144	32.89
YOLOv8s	OpenVINO	INT8	✅	11.2	0.7062	26.13
YOLOv8m	PyTorch	FP32	✅	49.7	0.737	201.44
YOLOv8m	OpenVINO	FP32	✅	99.1	0.7284	54.4
YOLOv8m	OpenVINO	INT8	✅	25.5	0.7268	30.76
YOLOv8l	PyTorch	FP32	✅	83.7	0.7769	385.46
YOLOv8l	OpenVINO	FP32	✅	167.0	0.7539	80.1
YOLOv8l	OpenVINO	INT8	✅	42.6	0.7508	52.25
YOLOv8x	PyTorch	FP32	✅	130.5	0.7759	609.4
YOLOv8x	OpenVINO	FP32	✅	260.6	0.7637	104.79
YOLOv8x	OpenVINO	INT8	✅	66.0	0.8077	64.96

Reproduce Our Results

To reproduce the Ultralytics benchmarks above on all export formats run this code:

from ultralytics import YOLO

# Load a YOLOv8n PyTorch model
model = YOLO("yolov8n.pt")

# Benchmark YOLOv8n speed and accuracy on the COCO8 dataset for all export formats
results = model.benchmarks(data="coco8.yaml")

# Benchmark YOLOv8n speed and accuracy on the COCO8 dataset for all export formats
yolo benchmark model=yolov8n.pt data=coco8.yaml

Conclusion

The benchmarking results clearly demonstrate the benefits of exporting the YOLOv8 model to the OpenVINO format. Across different models and hardware platforms, the OpenVINO format consistently outperforms other formats in terms of inference speed while maintaining comparable accuracy.

For the Intel® Data Center GPU Flex Series, the OpenVINO format was able to deliver inference speeds almost 10 times faster than the original PyTorch format. On the Xeon CPU, the OpenVINO format was twice as fast as the PyTorch format. The accuracy of the models remained nearly identical across the different formats.

The benchmarks underline the effectiveness of OpenVINO as a tool for deploying deep learning models. By converting models to the OpenVINO format, developers can achieve significant performance improvements, making it easier to deploy these models in real-world applications.

For more detailed information and instructions on using OpenVINO, refer to the official OpenVINO documentation.

FAQ

How do I export YOLOv8 models to OpenVINO format?

Exporting YOLOv8 models to the OpenVINO format can significantly enhance CPU speed and enable GPU and NPU accelerations on Intel hardware. To export, you can use either Python or CLI as shown below:

from ultralytics import YOLO

# Load a YOLOv8n PyTorch model
model = YOLO("yolov8n.pt")

# Export the model
model.export(format="openvino")  # creates 'yolov8n_openvino_model/'

# Export a YOLOv8n PyTorch model to OpenVINO format
yolo export model=yolov8n.pt format=openvino  # creates 'yolov8n_openvino_model/'

For more information, refer to the export formats documentation.

What are the benefits of using OpenVINO with YOLOv8 models?

Using Intel's OpenVINO toolkit with YOLOv8 models offers several benefits:

1. Performance: Achieve up to 3x speedup on CPU inference and leverage Intel GPUs and NPUs for acceleration.
2. Model Optimizer: Convert, optimize, and execute models from popular frameworks like PyTorch, TensorFlow, and ONNX.
3. Ease of Use: Over 80 tutorial notebooks are available to help users get started, including ones for YOLOv8.
4. Heterogeneous Execution: Deploy models on various Intel hardware with a unified API.

For detailed performance comparisons, visit our benchmarks section.

How can I run inference using a YOLOv8 model exported to OpenVINO?

After exporting a YOLOv8 model to OpenVINO format, you can run inference using Python or CLI:

from ultralytics import YOLO

# Load the exported OpenVINO model
ov_model = YOLO("yolov8n_openvino_model/")

# Run inference
results = ov_model("https://ultralytics.com/images/bus.jpg")

# Run inference with the exported model
yolo predict model=yolov8n_openvino_model source='https://ultralytics.com/images/bus.jpg'

Refer to our predict mode documentation for more details.

Why should I choose Ultralytics YOLOv8 over other models for OpenVINO export?

Ultralytics YOLOv8 is optimized for real-time object detection with high accuracy and speed. Specifically, when combined with OpenVINO, YOLOv8 provides:

• Up to 3x speedup on Intel CPUs
• Seamless deployment on Intel GPUs and NPUs
• Consistent and comparable accuracy across various export formats

For in-depth performance analysis, check our detailed YOLOv8 benchmarks on different hardware.

Can I benchmark YOLOv8 models on different formats such as PyTorch, ONNX, and OpenVINO?

Yes, you can benchmark YOLOv8 models in various formats including PyTorch, TorchScript, ONNX, and OpenVINO. Use the following code snippet to run benchmarks on your chosen dataset:

from ultralytics import YOLO

# Load a YOLOv8n PyTorch model
model = YOLO("yolov8n.pt")

# Benchmark YOLOv8n speed and [accuracy](https://www.ultralytics.com/glossary/accuracy) on the COCO8 dataset for all export formats
results = model.benchmarks(data="coco8.yaml")

# Benchmark YOLOv8n speed and accuracy on the COCO8 dataset for all export formats
yolo benchmark model=yolov8n.pt data=coco8.yaml

For detailed benchmark results, refer to our benchmarks section and export formats documentation.

📅 Created 1 year ago ✏️ Updated 2 months ago

Comments