开始使用

使用 Ultralytics YOLO26 和 Streamlit 应用进行实时推理

简介#

Streamlit 让你可以轻松构建和部署交互式 Web 应用。将它与 Ultralytics YOLO26 结合,即可直接在浏览器中进行实时目标检测和分析。YOLO26 的速度和准确率使其适用于安防、零售等应用中的实时视频流。



观看: 如何使用 Ultralytics YOLO26 和 Streamlit 构建实时推理应用 | 检测和分割 🚀
水产养殖畜牧业
使用 Ultralytics YOLO26 检测鱼类使用 Ultralytics YOLO26 检测动物
使用 Ultralytics YOLO26 检测鱼类使用 Ultralytics YOLO26 检测动物

实时推理的优势#

  • 实时目标检测:Streamlit 与 YOLO26 结合后,可以直接通过网络摄像头画面实时检测对象。这能够立即进行分析并提供洞察,因此非常适合需要即时反馈的应用。
  • 易于使用的部署方式:Streamlit 的交互式界面让你无需丰富的技术知识,也能轻松部署和使用应用。用户只需轻点一下即可开始实时推理,提升应用的易用性和可访问性。
  • 高效利用资源:YOLO26 的优化算法能以最少的计算资源实现高速处理。这种高效性让标准硬件也能流畅可靠地进行网络摄像头推理,使更多人能够使用先进的计算机视觉技术。

Streamlit 应用代码#

Ultralytics 安装

开始构建应用之前,请确保已安装 Ultralytics Python 软件包。

pip install ultralytics
使用 Streamlit 和 Ultralytics YOLO 进行推理
yolo solutions inference

yolo solutions inference model="path/to/model.pt"

yolo solutions help # list the available solution commands and arguments

这些命令会启动 Ultralytics 随附的默认 Streamlit 界面。

这将在你的默认网络浏览器中启动 Streamlit 应用。你会看到主标题、副标题,以及带有配置选项的侧边栏。选择所需的 YOLO26 模型,设置置信度和 NMS 阈值,然后点击“Start”按钮开始实时目标检测。

工作原理#

在底层,Streamlit 应用使用 Ultralytics solutions 模块创建交互式界面。开始推理后,应用会:

  1. 从网络摄像头、上传的视频文件或上传的图像中捕获帧
  2. 使用 YOLO26 模型处理每一帧
  3. 根据你指定的置信度和 IoU 阈值执行目标检测
  4. 实时显示原始帧和标注帧
  5. 如果选择启用,则开启目标跟踪

该应用提供简洁、易用的界面,其中包含用于调整模型参数以及随时开始或停止推理的控件。

结论#

按照本指南,你已经使用 Streamlit 和 Ultralytics YOLO26 创建了实时目标检测应用。该应用通过简单的界面在网络摄像头、视频或图像输入上运行 YOLO26,你可以随时停止视频流。

如需进一步增强功能,你可以考虑添加更多特性,例如录制视频流、保存标注帧,或集成其他计算机视觉库。

常见问题#

  • 使用 Streamlit 和 Ultralytics YOLO26 搭建实时目标检测应用非常简单。首先,使用以下命令确保已安装 Ultralytics Python 软件包:

    pip install ultralytics

    然后,你可以创建一个基本的 Streamlit 应用来运行实时推理:

    Streamlit 应用
    from ultralytics import solutions
    
    inf = solutions.Inference(
        model="yolo26n.pt",  # 你可以使用 Ultralytics 支持的任何模型,例如 YOLO26,或自定义训练的模型
    )
    
    inf.inference()
    
    # 确保使用命令 `streamlit run path/to/file.py` 运行文件

    如需了解实际搭建的更多详情,请参阅文档中的 Streamlit 应用代码部分。

  • 使用 Ultralytics YOLO26 和 Streamlit 进行实时目标检测有以下几项优势:

    • 实时检测:直接从网络摄像头画面中实现高准确率的实时目标检测。
    • 易用界面:Streamlit 的直观界面无需丰富的技术知识,即可轻松使用和部署。
    • 资源效率高:YOLO26 的优化算法能够以最少的计算资源实现高速处理。

    在实时推理的优势部分了解这些优势的更多信息。

  • 编写好集成 Ultralytics YOLO26 的 Streamlit 应用后,你可以运行以下命令进行部署:

    streamlit run path/to/file.py

    此命令会在默认 Web 浏览器中启动应用,让你选择 YOLO26 模型、设置置信度和 IoU 阈值,并通过简单的点击开始实时目标检测。

  • 使用 Streamlit 和 Ultralytics YOLO26 进行实时目标检测,可应用于多个领域:

    如需了解更深入的应用场景和示例,请查看 Ultralytics Solutions。

  • Ultralytics YOLO26 相较于 YOLOv5 和 R-CNN 系列检测器等早期模型,进行了多项改进:

    • 速度和精度更高:为实时应用带来更出色的性能。
    • 易于使用:界面和部署流程更加简化。
    • 资源效率高:经过优化,在计算需求极低的情况下实现更快速度。

评论26

提出问题、分享结果,或帮助其他人。

  1. j3dd4hGitHub

    the manual execution using rknn model (yolo predict model='./best_rknn_model' source='ppe.jpg') is working but got an error when running under Streamlit.

    TypeError: model='best_rknn_model/best-rk3588.rknn' is not a supported model format. Ultralytics supports: ('PyTorch', 'TorchScript', 'ONNX', 'OpenVINO', 'TensorRT', 'CoreML', 'TensorFlow SavedModel', 'TensorFlow GraphDef', 'TensorFlow Lite', 'TensorFlow Edge TPU', 'TensorFlow.js', 'PaddlePaddle', 'MNN', 'NCNN', 'IMX', 'RKNN') See https://docs.ultralytics.com/modes/predict for help.

    1
    1. RizwanMunawarGitHub

      @j3dd4h Hi, thanks for sharing the details. We currently don't support the rknn model, but you can adjust this check in code to make the application work for rknn models. We will add support for this soon.

  2. NNQHuongGitHub

    AttributeError: module 'torch' has no attribute '__version__' Traceback: File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\streamlit\runtime\scriptrunner\exec_code.py", line 88, in exec_func_with_error_handling result = func() ^^^^^^ File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 579, in code_to_exec exec(code, module.__dict__) File "C:\WasteSegregation\app.py", line 1, in <module> from ultralytics import solutions File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\ultralytics\__init__.py", line 11, in <module> from ultralytics.models import NAS, RTDETR, SAM, YOLO, FastSAM, YOLOWorld File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\ultralytics\models\__init__.py", line 3, in <module> from .fastsam import FastSAM File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\ultralytics\models\fastsam\__init__.py", line 3, in <module> from .model import FastSAM File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\ultralytics\models\fastsam\model.py", line 5, in <module> from ultralytics.engine.model import Model File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\ultralytics\engine\model.py", line 11, in <module> from ultralytics.cfg import TASK2DATA, get_cfg, get_save_dir File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\ultralytics\cfg\__init__.py", line 12, in <module> from ultralytics.utils import ( File "C:\Users\myname\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\ultralytics\utils\__init__.py", line 51, in <module> TORCH_VERSION = torch.__version__ ^^^^^^^^^^^^^^^^^ This is the error I encountered when using from ultralytics import solutions solutions.inference(model =r'C:path\to\my\model.pt') method (VSCode), can you please explain it to me? For further information, I've tried using the method for CLI and it works, but not with VSCode. I wonder if there's any method for CLI that allows me to use my custom trained model? It's YOLO11n-cls version. Thank you for reading this!

    1
    1. Glenn JocherUltralytics

      @NNQHuong the error you're encountering suggests a possible issue with your PyTorch installation. Please verify that PyTorch is properly installed and compatible with your Python version by running:

      import torch
      print(torch.__version__)

      If the error persists, ensure you're using the latest version of PyTorch and Ultralytics with:

      pip install --upgrade torch ultralytics

      Regarding your CLI query, you can use your custom-trained YOLO11 model by specifying the model path directly, like so:

      yolo predict model=path/to/your/model.pt source=your_input

      For detailed examples, check the documentation here. Let me know if additional assistance is required!

      2
    2. RizwanMunawarGitHub

      Hi @NNQHuong, thank you for sharing the error log! It appears that the issue might be related to your PyTorch version. You may need to upgrade or downgrade it to ensure everything works correctly.

      However, if this still doesn't resolve the issue, please share your code here for further testing.

      Thanks, Ultralytics Team!

      1
    3. NNQHuongGitHub

      Thanks for your infomative reply, it worked smoothly! I'm kind of new to this field anyway, about 6 months , I bet you can see it in my words haha, so thanks for your patience too!

      1
  3. Tech-WattGitHub

    Is it posible to use different camera index ?. It picks up the default computer webcam and there's no means of changing it.

    1. Glenn JocherUltralytics(已编辑)

      The Inference class has no source argument: selecting "webcam" in the sidebar always opens camera index 0 (self.vid_file_name = 0 in ultralytics/solutions/streamlit_inference.py). To use a different camera, make it your system's default camera or change that index in the file:

      inf = solutions.Inference(model="yolo26n.pt")
      inf.inference()

      Run it with streamlit run <file-name.py>.

  4. VectorvinGitHub

    streamlit 这个解决方案可以使用tensorRT格式或者onnx格式吗?似乎程序被写死了

    1. Glenn JocherUltralytics(已编辑)

      @Vectorvin 您好!Streamlit 方案现在也支持导出的模型格式:除了 .pt,还可以把 .onnx、.torchscript、.mlpackage、.engine(TensorRT)、OpenVINO 或 RKNN 模型路径传给 solutions.Inference(model="yolo26n.engine")。导出步骤请参考导出文档。需要企业级部署支持请访问ultralytics.com/license。

  5. sathyanarayanaRPGitHub

    how does the streamlit gets is input , is it possible to push frames or is it possible to integrate machine vision cameras that does not support RTSP protocals. i need comprehensive knowledge how does it work as a streaming device

    1. Glenn JocherUltralytics

      @sathyanarayanaRP streamlit can receive input directly from a webcam or video file. For cameras that don't support RTSP, you can use USB or other compatible connections. The Streamlit app processes frames in real-time, allowing for seamless integration with various camera types. For more details, please refer to the Streamlit documentation.

  6. Cheikh-AhmedandGitHub

    hey!

    Once I run the app my Laptop camera turns on, instead of My webcam. How to use my external webcam instead of my LAPTOP camera?

    Best Regards!

    1. Glenn JocherUltralytics(已编辑)

      @Cheikh-Ahmedand the Streamlit app has no source argument: choosing "webcam" in the sidebar always opens camera index 0 (self.vid_file_name = 0 in ultralytics/solutions/streamlit_inference.py). To use your external webcam, make it the system's default camera, or change that index to 1 (or the index of your webcam) in the file.

      Run the app with yolo solutions inference, or with streamlit run <file-name.py> for your own script.

    2. Cheikh-AhmedandGitHub

      Hey ! Thanks lot for your response, but the code couldn't work with streamlit code. I pasted the following code on my Laptop anconda PowerShell Prompt:

      yolo streamlit-predict solutions.inference(source=1)

      Still, my Webcam can't run, but my Laptop 's Camera.

      Also, I tried the below code in jupyter notebook, but the problem still persists.

      from ultralytics import solutions

      solutions.inference() solutions.inference(source=1)

      Make sure to run the file using command streamlit run <file-name.py>

      Any help either from the Respectful Staff or others would be appreciated!!!

    3. Glenn JocherUltralytics(已编辑)

      @Cheikh-Ahmedand the Streamlit app has no source argument: choosing "webcam" in the sidebar always opens camera index 0 (self.vid_file_name = 0 in ultralytics/solutions/streamlit_inference.py). To use an external webcam, set it as your system's default camera or change that index in the file. Run the app with yolo solutions inference, or streamlit run <file-name.py> for your own script. If issues persist, please check your system's camera settings.

    1. Glenn JocherUltralytics(已编辑)

      @AdadAlShabab you can use the following code to run the Streamlit application for live inference:

      from ultralytics import solutions
      
      inf = solutions.Inference(model="yolo26n.pt")
      inf.inference()

      Run the file with the command streamlit run <file-name.py>. For more details, please refer to the Streamlit Live Inference Guide.

  7. ghostGitHub

    Thank you for your quick and kind response and the problem is sorted. But there is another issue while switching model to different task like pose,seg or cls and getting error of open cv not having func()

    1. Glenn JocherUltralytics

      @Charan2128 i'm glad to hear the initial issue is resolved. Regarding the error with OpenCV when switching tasks (pose, seg, cls), please ensure you have the latest versions of both the Ultralytics package and OpenCV. You can update them using:

      pip install --upgrade ultralytics opencv-python

      If the issue persists, please provide more details about the error message and the specific function that is causing the problem. This will help us diagnose and address the issue more effectively. For further guidance, you can also refer to our Streamlit Live Inference Guide.

    2. ghostGitHub

      That was a Quick Replay and Quick fix ,Thanks for your kind support on my issues. The problem is fixed now.

    3. Glenn JocherUltralytics

      @Charan2128 i'm glad to hear the initial issue is resolved. For the new issue with switching tasks and encountering OpenCV errors, please ensure you have the latest versions of both the Ultralytics package and OpenCV installed. You can update them using pip install --upgrade ultralytics opencv-python. If the problem persists, please provide the specific error message and function causing the issue so we can assist you further. For more detailed guidance, you can refer to our Streamlit Live Inference Guide.

  8. ghostGitHub

    AttributeError: module 'ultralytics.solutions' has no attribute 'inference' Traceback: File "C:\Users\DELL\AppData\Local\Programs\Python\Python312\Lib\site-packages\streamlit\runtime\scriptrunner\exec_code.py", line 75, in exec_func_with_error_handling result = func() ^^^^^^ File "C:\Users\DELL\AppData\Local\Programs\Python\Python312\Lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 574, in code_to_exec exec(code, module.dict) File "D:\YOLO\infer.py", line 3, in solutions.inference() ^^^^^^^^^^^^^^^^^^^ This is the error im getting while running this and i have installed necessary packages.

    1. Paula DerrengerUltralytics(已编辑)

      @Charan2128 it looks like you're encountering an AttributeError because solutions.inference is not an attribute of the ultralytics.solutions module. Please ensure you are using the latest version of the Ultralytics package by running pip install --upgrade ultralytics. Then launch the Streamlit app directly from the CLI with yolo solutions inference, or from Python with solutions.Inference(model="yolo26n.pt").inference() run via streamlit run <file-name.py>. For more details, you can refer to our Streamlit Live Inference Guide.