Ultralytics
zh/modes/export/ #9406
giscus[bot]
bot
Replies: 13 comments 44 replies
-
|
how to export to ncnn? the official example is wrong. will get SegmentError. |
Beta Was this translation helpful? Give feedback.
-
|
请问如何将pt模型修改后，如何导出为pt文件？我不设置format参数，发现默认导出的是torchscript格式 |
Beta Was this translation helpful? Give feedback.
-
|
您好！关于如何将 import torch
from ultralytics import YOLO
# 加载模型
model = YOLO('path/to/your/model.pt') # 加载您的.pt模型
# 进行您的模型修改
# model.xxx = ...
# 保存模型
torch.save(model.state_dict(), 'path/to/save/your_modified_model.pt')如果您希望直接从YOLOv8模型进行修改并导出为TorchScript（默认格式），则可以使用以下方法： from ultralytics import YOLO
# 加载模型
model = YOLO('path/to/your/model.pt')
# 导出为TorchScript，默认不需要设置format参数
model.export('path/to/save/your_model.torchscript')对于NCNN格式的导出问题，我们已经注意到了相关的SegmentError问题，在修复更新前您可以尝试其他格式的导出。如果有进一步的问题或者更新，我们会及时通知社区的。希望这能帮助到您！😊 |
Beta Was this translation helpful? Give feedback.
-
|
您好！关于如何导出修改后的 from ultralytics import YOLO
# 假设您已经加载并可能对模型进行了某种形式的修改
model = YOLO('path/to/your/modified_model.pt')
# 直接保存修改后的模型
model.save('path/to/your/exported_model.pt')📝 注意：在当前 希望这有助于您完成模型的导出需求！如果有进一步的问题或者需要关于模型转化为其他格式的帮助（比如 |
Beta Was this translation helpful? Give feedback.
-
|
感谢回复！我使用的模型是RTDETR，调用model.save()方法会出现如下错误信息：AttributeError: 'RTDETR' object has no attribute 'save'。 |
Beta Was this translation helpful? Give feedback.
-
|
@Ameridru 👋 感谢您的反馈！对于 RTDETR 模型和 import torch
# 假设 'model' 是您已经加载或训练的 RTDETR 模型
torch.save(model.state_dict(), 'path/to/your_rtDETR_model.pt')📘 这种方法会保存模型的权重，而不是整个模型架构。如果您需要重新加载模型，您需要首先创建模型架构的实例，然后加载状态字典。 请注意，如果 希望这有助于解决您遇到的问题！如果还有其他疑问，欢迎随时提问。 |
Beta Was this translation helpful? Give feedback.
-
|
希望可以指定导出目录 |
Beta Was this translation helpful? Give feedback.
-
|
@2213766112 您好！👋 可以通过添加 yolo export model=yolov8n.pt format=onnx project=exports name=my_model这将会创建 |
Beta Was this translation helpful? Give feedback.
-
|
执行命令 yolo export model=E:\Geetest4icon\YOLOV8_分类_标题\YOLOV8_训练\weights\best.pt project=E:\Geetest4icon\YOLOV8_分类_标题 name=YOLOV8_导出 format=openvino batch=1 导出路径会乱码 E:\Geetest4icon\YOLOV8_鍒嗙被_鏍囬\YOLOV8_璁粌\weights\best_openvino_model 而且 导出的路径并不是设置的project下的name文件夹内 执行命令 yolo export model=E:\Geetest4icon\YOLOV8_分类_标题\YOLOV8_训练\weights\best.pt project=E:\Geetest4icon\YOLOV8_分类_标题 name=YOLOV8_导出 format=onnx 跟上面命令不同的是 导出的是onnx 执行这个命令不会乱码 但是依旧导出路径不对 此命令会在model所在的相同目录下输出best.onnx文件 求解决 拜托了 |
Beta Was this translation helpful? Give feedback.
-
|
你好！👋 对于导出路径乱码的问题，这很可能是因为路径中包含了非ASCII字符。对于Windows系统，建议尝试将路径或文件名转换为英文或避免使用特殊字符。此外，请确保您使用的是最新版本的 至于导出路径不正确的问题，这可能是因为导出命令的某些参数设置不正确。请确保 yolo export model=path/to/best.pt project=path/to/export/directory name=model_export format=onnx请注意， 如果仍然遇到问题，请确保您的 |
Beta Was this translation helpful? Give feedback.
-
|
你好！在我将pt转换为engine时，出现 |
Beta Was this translation helpful? Give feedback.
-
|
你好！遇到这个 yolo export model=yolov8n.pt format=engine如果问题依旧，请确保所有相关的依赖都已正确安装，并尝试重新安装TensorRT。如果仍然无法解决，欢迎通过GitHub issue提供更多信息，如TensorRT版本和完整的错误日志，我们乐意进一步协助！👍 |
Beta Was this translation helpful? Give feedback.
-
|
我还是没有解决问题 |
Beta Was this translation helpful? Give feedback.
-
|
@Godyan0804 你好！看起来你在尝试将
解决建议：
如果问题仍然存在，请尝试清晰地报告你的问题以及任何错误消息到 Ultralytics YOLOv8的 GitHub issues 页面，包括你尝试了哪些步骤以及详细的环境配置。这样，社区和开发者可以提供更具体的帮助。希望这些建议能帮到你！🙌 |
Beta Was this translation helpful? Give feedback.
-
|
执行命令 yolo export model=E:\Geetest4icon\YOLOV8_分类_标题\YOLOV8_训练\weights\best.pt project=E:\Geetest4icon\YOLOV8_分类_标题 name=YOLOV8_导出 format=openvino batch=1 导出路径会乱码 E:\Geetest4icon\YOLOV8_鍒嗙被_鏍囬\YOLOV8_璁粌\weights\best_openvino_model 而且 导出的路径并不是设置的project下的name文件夹内 执行命令 yolo export model=E:\Geetest4icon\YOLOV8_分类_标题\YOLOV8_训练\weights\best.pt project=E:\Geetest4icon\YOLOV8_分类_标题 name=YOLOV8_导出 format=onnx 跟上面命令不同的是 导出的是onnx 执行这个命令不会乱码 但是依旧导出路径不对 此命令会在model所在的相同目录下输出best.onnx文件 求解决 拜托了 |
Beta Was this translation helpful? Give feedback.
-
|
@2213766112 您好！感谢您的耐心等待和详细说明。看起来您在导出模型时遇到了两个问题：
关于路径乱码，这可能是因为命令行界面（CLI）或系统对特定字符集的支持问题。尝试使用英文路径和文件名，避免使用中文或特殊字符，以确保路径能被正确解析。 至于导出路径不正确的问题， 作为参考，以下是一个简单的命令示例，用于导出模型并避免路径问题（请根据需要调整模型和输出格式）: yolo export model=best.pt format=onnx执行后，导出的 希望这能帮助您解决问题！如果还有其他疑问，欢迎继续提问。🙂 |
Beta Was this translation helpful? Give feedback.
-
|
请问 后续的export命令 会考虑支持project 和 name 参数吗 |
Beta Was this translation helpful? Give feedback.
-
|
@2213766112 您好！非常感谢您的提议和反馈🙂。 目前 我会将您的反馈提交给我们的开发团队，以便在未来的版本中考虑此功能。您的建议对于改进YOLOv8的功能和用户体验非常有价值。再次感谢您的积极参与和宝贵意见! 🚀 如果您有任何其他问题或建议，请随时告知。 |
Beta Was this translation helpful? Give feedback.
-
|
F:\conda\python.exe F:\yolov8\1.py PyTorch: starting from 'yolov8n.pt' with input shape (1, 3, 640, 640) BCHW and output shape(s) (1, 84, 8400) (6.2 MB) ONNX: starting export with onnx 1.16.0 opset 17... TensorRT: starting export with TensorRT 10.0.1... 进程已结束，退出代码为 1 |
Beta Was this translation helpful? Give feedback.
-
|
@Godyan0804 it looks like you encountered an error related to the To resolve this, ensure that you are using a compatible version of both the NVIDIA drivers and the TensorRT library that matches the requirements or recommendations of the specific version of the Ultralytics YOLOv8. Here's a checklist you might want to go through:
If the issue persists after checking these, consider re-installing TensorRT or trying on another machine to rule out environment-specific issues. You can also share more details here or on Ultralytics' GitHub discussions for further community support. |
Beta Was this translation helpful? Give feedback.
-
|
我想要把模型保存为onnx格式，并指定路径，我的代码是 但是发生报错，报错的内容是： 请问这个问题的原因是什么 |
Beta Was this translation helpful? Give feedback.
-
|
How do I convert an existing pt model to onnx? |
Beta Was this translation helpful? Give feedback.
-
|
Sure! To convert an existing Python Codefrom ultralytics import YOLO
# Load your pretrained model
model = YOLO('path/to/your/model.pt')
# Export the model to ONNX format
model.export(format='onnx')CLI Commandyolo export model=path/to/your/model.pt format=onnxThis will create an |
Beta Was this translation helpful? Give feedback.
-
|
当我导出onnx时使用了dynamic参数时，导出的模型无法使用 yolo val 来验证，会得到以下的错误： |
Beta Was this translation helpful? Give feedback.
-
|
经过测试，是dynamic与half不兼容导致的 |
Beta Was this translation helpful? Give feedback.
-
|
确实，当使用 作为解决方案，你可以尝试以下几种方法：
如果仍有问题，建议在 Ultralytics GitHub issues 页面提出，我们的开发者社区和团队成员将提供进一步的帮助！ # 试试这样导出模型
yolo export model=your_model.pt format=onnx dynamic=True half=False希望这可以帮助到你！如果有进一步的问题或需求，请随时回复。 😊 |
Beta Was this translation helpful? Give feedback.
-
|
您好，使用.pt和导出为.onnx进行yolov8模型的图片中目标检测时，参数设置一致，但是最终检测框数量，位置不完全一致。比如.onnx比.pt多画一个或者检测到.pt检测不出来的类别，请问是什么问题导致的呢？ |
Beta Was this translation helpful? Give feedback.
-
|
@nacenace 您好， 感谢您提出这个问题！在导出模型时遇到检测结果不一致的情况确实令人困惑。以下是一些可能的原因和解决方法：
请尝试以上建议，如果问题仍然存在，请提供更多详细信息和代码示例，以便我们进一步调查。 希望这些信息对您有所帮助！如果有任何其他问题，请随时告知。 祝好， |
Beta Was this translation helpful? Give feedback.
-
|
你好，我和你有相同的问题，问题解决了吗 |
Beta Was this translation helpful? Give feedback.
-
|
没有 |
Beta Was this translation helpful? Give feedback.
-
|
您好，感谢您的反馈。导出为不同格式（如ONNX）时，检测结果可能会有所不同，通常是由于模型转换过程中精度损失或推理引擎差异导致的。建议您确保使用最新版本的Ultralytics和相关库，并尝试调整导出参数，如 |
Beta Was this translation helpful? Give feedback.
-
|
当我想把模型转换为onnx int8类型，命令行输入：yolo export model=yolov8n.pt format=onnx int8=True，但是得到的onnx文件依然是float32位，按照提示输入：yolo export model=yolov8n.pt format=onnx int8=True data=coco8.yaml；得到的onnx依然还是float32位，想问下什么原因 |
Beta Was this translation helpful? Give feedback.
-
|
@Lcw666666 感谢您的反馈。请确保您使用的是最新版本的Ultralytics库，因为我们不断更新和修复问题。导出为INT8量化的ONNX模型时，可能需要确保您的环境支持INT8量化，并且所有依赖项已正确安装。如果问题仍然存在，请尝试在不同的环境中运行或检查是否有任何错误日志输出。更多详细信息可以参考我们的导出文档。 |
Beta Was this translation helpful? Give feedback.
-
|
我使用pt的模型在yolo tracker中可以用，后来将其导出为onnx。导出没报错，但是yolo tracker中无法使用了。可能的原因是什么呢？ |
Beta Was this translation helpful? Give feedback.
-
|
You can set the meta info outside Ultralytics by modifying the exported ONNX model's metadata manually. However, for a more integrated solution, you might need to fork the repo and update the code to ensure the necessary attributes are initialized correctly for Triton. |
Beta Was this translation helpful? Give feedback.
-
|
actually there's meta info in onnx model, but the meta info of model won't be loaded when using remote triton server. |
Beta Was this translation helpful? Give feedback.
-
|
Adding the metainfo which is extracted from onnx metainfo in AutoBackend.py when initialing triton inference server, everything is ok... |
Beta Was this translation helpful? Give feedback.
-
|
I have made it as an PR which try to deliver metadata when initialize model hosted on triton without metadata. Hope this can help others. Thanks. |
Beta Was this translation helpful? Give feedback.
-
|
@quanqigu thank you for your contribution! Your PR to deliver metadata when initializing models hosted on Triton without metadata is greatly appreciated. This will indeed help others facing similar issues. |
Beta Was this translation helpful? Give feedback.
-
|
请问这个如何解决 |
Beta Was this translation helpful? Give feedback.
-
|
@2972749844 to resolve the ONNX export issue with adaptive pooling, you might need to modify the model architecture to use a fixed output size. For further assistance, consider checking the PyTorch GitHub for updates or submitting a request there. |
Beta Was this translation helpful? Give feedback.
-
Why is the onnx converted with this code completely different from the original pt parameters, and the output of the same image is also different |
Beta Was this translation helpful? Give feedback.
-
|
@ganganngannn the difference in outputs between the original |
Beta Was this translation helpful? Give feedback.
-
|
I only used the original training and conversion onnx configuration, here is my code： training：conversion:Where did I go wrong？The parameters for the first 96 * 3 * 3 * 3 convolution in the PT model are（float16）： The parameters for the first 96 * 3 * 3 * 3 convolution in the ONNX model are（float32）： |
Beta Was this translation helpful? Give feedback.
-
|
I found that 'half=True' or 'int=True' are not useful, the ONNX parameters are both float32 and the same. Is there anything wrong with that? |
Beta Was this translation helpful? Give feedback.
-
|
@ganganngannn the issue likely arises because the |
Beta Was this translation helpful? Give feedback.
-
zh/modes/export/
逐步指导如何将YOLOv8 模型导出为各种格式，如ONNX,TensorRT,CoreML 等，以便部署。立即探索
https://docs.ultralytics.com/zh/modes/export/
Beta Was this translation helpful? Give feedback.
All reactions