I am getting the error message "model... .xml is not a supported model format" when it executes this python code: "results=model(imageFrame, conf=0.2)".
I am trying to run inference on an OpenVINO exported model with Ultralytics API following this page: https://docs.openvino.ai/2024/notebooks/yolov10-optimization-with-output.html
This is part of my python program:
import cv2
from ultralytics import YOLO
pt_model = pt_model_path + 'best.pt'
onnx_model = onnx_model_path + 'best.onnx'
ov_model = openvino_model_path + 'best.xml'
model = YOLO(ov_model, task="detect") # <-- WORKS WELL WITH ALL MODELS
cap = cv2.VideoCapture(video_file)
...
while...
imageFrame = cap.read()
...
results=model(imageFrame, conf=0.2) # <-- ERROR WHEN USING OPENVINO MODEL
...
annotated_frame = results[0].plot() # Visualize the results on the frame
cv2.imshow("Y10 inference", annotated_frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
I am using a Windows 11 machine with CUDA (NVIDIA RTX 4060) and, I have to believe I am using YoloV10 because somehow, I understand it is now integrated in Ultralytics. These are the steps I performed:
- Installed Python 3.10.4 globally (I uninstalled all other versions globally first)
- Created (and then activated) the python virtual environment "y_env" where I ran the following "pip" commands:
> pip install ultralytics # Gets me version 8.2.58
# I downloaded the requirements.txt file from https://github.com/THU-MIG/yolov10/tree/main
> pip install -r requirements.txt
> pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
- I created the custom model by training with "yoloV10n.pt" using my GPU (it took nearly two minutes to finish) as follows:
> yolo task=detect mode=train model=yolov10n.pt data=<path>\data.yaml epochs=100 imgsz=416 plots=True device=0
- When finished training, I first tested my python program with the [<path>\runs\detect\train9\weights\best.pt] PyTorch file and it worked well displaying the images with bounding boxes.
- I then exported the PyTorch file to ONNX and OpenVINO formats as follows:
> yolo export model=<path>\runs\detect\train9\weights\best.pt format=openvino
> yolo export model=<path>\runs\detect\train9\weights\best.pt format=onnx
- I tested my python program with the [<path>\runs\detect\train9\weights\best.onnx] ONNX file and it worked well displaying the images with bounding boxes.
I also got the same error after performing the above steps with the "YoloV10" that I cloned from these two GitHub links:
- https://github.com/THU-MIG/yolov10.git (which gives Ultralytics 8.1.34)
- https://github.com/ultralytics/ultralytics.git@main (which gives Ultralytics 8.2.58).
Questions:
- Why am I getting the "model... .xml is not a supported model format" error when running the program with the XML OpenVINO model?
- If you find me a solution to this problem, could you please detail me all the steps I need to fix it? I will really appreciate it.
Thank you in advance and eagerly waiting for your reply,
Manuel Hernandez
