YOLO Models

This page illustrates how to prepare models from the YOLO family for integration with the Intel® DL Streamer pipeline.

Note

The instructions provided below are comprehensive, but for convenience, it is recommended to use the download_public_models.sh script. This script will download all supported Yolo models and perform the necessary conversions automatically.

See download_public_models for more information.

1. Setup

The instructions assume Intel® DL Streamer framework is installed on the local system along with Intel® OpenVINO™ model downloader and converter tools, as described here: Tutorial.

For YOLOv5, YOLOv8, YOLOv9, YOLOv10, YOLO11 models it is also necessary to install the Ultralytics Python package:

pip install ultralytics

2. YOLOv8, YOLOv9, YOLOv10, YOLO11

Python script converting the recent Ultralytics models to Intel® OpenVINO™ format (replace model_name and model_type with the relevant models and types values):

Example for YOLO11 model:

from ultralytics import YOLO
import openvino, sys, shutil, os

model_name = 'yolo11s'
model_type = 'yolo_v11'
weights = model_name + '.pt'
model = YOLO(weights)
model.info()

converted_path = model.export(format='openvino')
converted_model = converted_path + '/' + model_name + '.xml'

core = openvino.Core()

ov_model = core.read_model(model=converted_model)
if model_type in ["YOLOv8-SEG", "yolo_v11_seg"]:
    ov_model.output(0).set_names({"boxes"})
    ov_model.output(1).set_names({"masks"})
ov_model.set_rt_info(model_type, ['model_info', 'model_type'])

openvino.save_model(ov_model, './FP32/' + model_name + '.xml', compress_to_fp16=False)
openvino.save_model(ov_model, './FP16/' + model_name + '.xml', compress_to_fp16=True)

shutil.rmtree(converted_path)
os.remove(f"{model_name}.pt")

3. YOLOv7

Model preparation:

git clone https://github.com/WongKinYiu/yolov7.git
cd yolov7
python3 export.py --weights yolov7.pt --grid --dynamic-batch
# Convert the model to OpenVINO format FP32 precision
ovc yolov7.onnx --compress_to_fp16=False
# Convert the model to OpenVINO format FP16 precision
ovc yolov7.onnx

4. YOLOv5 latest version

Model preparation enabling the dynamic batch size:

from ultralytics import YOLO
from openvino.runtime import Core
from openvino.runtime import save_model
model = YOLO("yolov5s.pt")
model.info()
model.export(format='openvino', dynamic=True)  # creates 'yolov5su_openvino_model/'
core = Core()
model = core.read_model("yolov5su_openvino_model/yolov5su.xml")
model.reshape([-1, 3, 640, 640])
save_model(model, "yolov5su.xml")

5. YOLOv5 older versions

Model preparation of YOLOv5 7.0 from Ultralytics involves two steps. First, convert the PyTorch model to Intel® OpenVINO™ format :

git clone https://github.com/ultralytics/yolov5
cd yolov5
wget https://github.com/ultralytics/yolov5/releases/download/v7.0/yolov5s.pt
python3 export.py --weights yolov5s.pt --include openvino --dynamic

Then, reshape the model to enable the dynamic batch size and keep other dimensions fixed:

from openvino.runtime import Core
from openvino.runtime import save_model
core = Core()
model = core.read_model("yolov5s_openvino_model/yolov5s.xml")
model.reshape([-1, 3, 640, 640])
save_model(model, "yolov5s.xml")

6. YOLOX

Intel® OpenVINO™ version of the model can be obtained from the ONNX file:

wget https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_s.onnx
ovc yolox_s.onnx --compress_to_fp16=False

7. Model usage

See Samples for detailed examples of Intel® DL Streamer pipelines using different Yolo models.