zh/integrations/onnx/

[giscus[bot]]

zh/integrations/onnx/

了解如何将YOLO11 模型导出为ONNX 格式，以便在各种平台上灵活部署并提高性能。

https://docs.ultralytics.com/zh/integrations/onnx/

[xiao5676wen]

我想知道，导出onnx后怎样用onnx实现这段YOLO代码功能：
from ultralytics import YOLO
import torch
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = YOLO('E:\yolo11_go\runs\detect\train2\weights\best.pt')

results = model.track('1.mp4', conf=0.75, show_conf=False, line_width=1, device=device, persist=True)
for result in results:
if result.boxes.id is not None:
track_ids_this = result.boxes.id
track_ids_this = max(track_ids_this.tolist())
print(track_ids_this)

[glenn-jocher]

@xiao5676wen to implement the provided YOLO tracking functionality with an ONNX-exported YOLO model, you will need to use ONNX Runtime or similar frameworks for inference since the track mode is not natively supported in ONNX. You can process the inference results manually to replicate the tracking logic. For guidance on exporting YOLO11 models to ONNX and deploying them with ONNX Runtime, please refer to our documentation here: ONNX Export for YOLO11 Models. Let us know if you need further clarification!

[UltralyticsAssistant]

👋 Hello, thank you for your interest in Ultralytics 🚀! We recommend exploring the resources in our Documentation to learn more about YOLO and its integrations, including ONNX export, which provides flexibility for deployment across various platforms.

If this is a 🐛 Bug Report related to ONNX integration or export, please provide a minimum reproducible example to assist us in resolving it efficiently.

If this is a ❓ Question about custom ONNX usage or deployment, kindly provide additional details such as the YOLO model version, specific export commands used, and any errors or logs encountered during the process.

For community support, you can also join our vibrant Ultralytics network:

• Real-time discussions on Discord 🎧
• Engage in threaded conversations on Discourse 💬
• Or share insights via our Subreddit 📚

Upgrade

Ensure you’re using the latest ultralytics library version by running:

pip install -U ultralytics

This ensures compatibility and access to the newest features. Check out our requirements file for dependencies, and confirm you’re operating on Python>=3.8 and PyTorch>=1.8.

Environments

YOLO models, including ONNX-related workflows, can be run in these verified setups:

• Notebooks with free GPU:
• Google Cloud Deep Learning VM (Quickstart Guide)
• Amazon Deep Learning AMI (Quickstart Guide)
• Docker Image (Quickstart Guide)

Status

If this badge shows green, all Ultralytics CI tests are passing, confirming compatibility with ONNX and other modes across platforms.

This is an automated response to help guide you 🔍. An Ultralytics engineer will review your query and provide additional assistance soon.

[xiao5676wen]

我想知道，导出onnx后怎样用onnx实现这段YOLO代码功能：
from ultralytics import YOLO
import torch
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = YOLO('E:\yolo11_go\runs\detect\train2\weights\best.pt')

results = model.track('1.mp4', conf=0.75, show_conf=False, line_width=1, device=device, persist=True)
for result in results:
if result.boxes.id is not None:
track_ids_this = result.boxes.id
track_ids_this = max(track_ids_this.tolist())
print(track_ids_this)

[glenn-jocher]

To use the functionality of your original YOLO code with an ONNX model, you can leverage onnxruntime for inference as ONNX models are not natively executable with the YOLO class. A simple example involves loading the ONNX model with ONNX Runtime, preparing the input, and processing the output. For tracking, you would need to implement custom logic as ONNX does not directly support the track method. Refer to ONNX Runtime documentation for detailed guidance.