开始使用

使用 Ultralytics YOLO26 进行分析

使用 Ultralytics YOLO26 进行分析,可将对象检测和跟踪结果转换为实时图表,让你逐帧查看视频中对象数量的变化。本指南介绍四种数据可视化类型——折线图、条形图、饼图和面积图,并通过通用的 Python 和 CLI 示例演示如何切换图表类型。



观看: 如何使用 Ultralytics 生成分析图表 | 折线图、条形图、面积图和饼图

可视化示例#

折线图条形图饼图
用于对象跟踪的 YOLO 分析折线图用于检测数量统计的 YOLO 分析条形图用于类别分布的 YOLO 分析饼图

为什么要将检测数据可视化?#

  • 折线图非常适合跟踪短期和长期变化,也适合比较多个组在同一时期内的变化。
  • 条形图适合比较不同类别的数量,以及展示类别与其数值之间的关系。
  • 饼图适用于呈现各类别所占比例,以及展示整体中的各个部分。
  • 面积图会填充折线图下方的区域,让你一目了然地查看各类别对象数量随时间的变化。

生成分析图表#

将视频传递给 Analytics 解决方案,并使用 analytics_type 选择图表。该解决方案会对每一帧执行检测和跟踪,并渲染默认大小为 1280×720 的图表,你可以直接将图表写入输出视频。只需一个参数,即可在 "line"、"bar"、"pie" 和 "area" 之间切换。

使用 Ultralytics YOLO 进行分析
yolo solutions analytics show=True

# Pass the source
yolo solutions analytics source="path/to/video.mp4"

# Generate the pie chart
yolo solutions analytics analytics_type="pie" show=True

# Generate the bar plots
yolo solutions analytics analytics_type="bar" show=True

# Generate the area plots
yolo solutions analytics analytics_type="area" show=True

Analytics 参数#

以下表格列出了 Analytics 的参数:

参数类型默认值说明
modelstrNoneUltralytics YOLO 模型文件的路径。
analytics_typestr'line'图表类型,即 line、bar、area 或 pie。
figsizetuple[float, float](12.8, 7.2)Analytics 图表的图幅尺寸(英寸);(12.8, 7.2) 会渲染 1280×720 的帧。
line_widthint2解决方案绘制的框、关键点和计数的线条粗细。
verboseboolTrue启用逐帧日志,记录输入形状、类别计数和处理速度。跟踪调用本身始终不会输出日志。

你也可以在 Analytics 解决方案中使用 track 参数。

参数类型默认值说明
trackerstr'tracktrack.yaml'指定要使用的跟踪算法。内置选项:botsort.yaml、bytetrack.yaml、ocsort.yaml、deepocsort.yaml、fasttrack.yaml、tracktrack.yaml。
conffloat0.25设置检测的置信度阈值;较低的值允许跟踪更多物体,但可能会包含误检。
ioufloat0.5设置用于筛除重叠检测结果的交并比 (IoU) 阈值。
classeslistNone按类别索引筛选结果。例如,classes=[0, 2, 3] 只跟踪指定类别。
devicestrNone指定推理设备(例如 cpu、cuda:0 或 0)。用户可以选择 CPU、特定 GPU 或其他计算设备来运行模型。
imgszint or tuple640目标推理图像尺寸:正方形图像使用整数,或使用 (height, width) 对。图像尺寸会调整为模型步幅的整数倍。
max_detint300跟踪前每帧保留的最大检测数量,用于限制解决方案在密集场景中处理的物体数量。
quantizeint or strNone推理精度:对于非 CPU 上运行的 PyTorch 和 TorchScript 模型,使用 16 (FP16) 或 32/unset (FP32) 进行选择;其他格式使用其模型文件和运行时所选的精度。此参数替代已弃用的 half 参数。

此外,还支持以下可视化参数:

参数类型默认值说明
showboolFalse如果 True,则在窗口中显示带标注的图像或视频。适用于在开发或测试期间即时查看结果。

结论#

了解何时以及如何使用不同类型的可视化图表,对于有效的数据分析至关重要。折线图、柱状图、饼图和面积图是基本工具,可帮助你更清晰、更有效地讲述数据所呈现的信息。Ultralytics YOLO26 Analytics 解决方案提供了一种精简的方法,可根据目标检测和跟踪结果生成这些可视化图表,让你更轻松地从视觉数据中提取有意义的见解。

常见问题#

  • 要使用 Ultralytics YOLO26 Analytics 创建折线图,请按以下步骤操作:

    1. 加载 YOLO26 模型并打开视频文件。
    2. 使用 analytics_type="line" 初始化 Analytics 类。
    3. 遍历视频帧,并在每一帧调用解决方案,用对象数量等数据更新折线图。
    4. 将 results.plot_im 写入输出视频,以保存图表。

    你可以从上面的 Python 示例开始着手——该示例已包含完整的帧循环,而折线图是默认的 analytics_type。

  • 使用 Ultralytics YOLO26 创建条形图有以下好处:

    1. 实时数据可视化:将目标检测结果整合到柱状图中,以动态更新图表。
    2. 易于使用:简单的 API 和函数让数据实现和可视化变得轻松直接。
    3. 可定制:自定义标题、标签、颜色等,以满足你的具体需求。
    4. 效率:高效处理大量数据,并在视频处理期间实时更新图表。

    要生成条形图,请在上面的 Python 示例中设置 analytics_type="bar"——其余帧循环保持不变。有关预览,请参阅可视化示例部分。

  • Ultralytics YOLO26 非常适合创建饼图,原因如下:

    1. 与对象检测集成:将对象检测结果直接整合到饼图中,快速获取见解。
    2. 易用的 API:只需少量代码即可轻松设置和使用。
    3. 可定制:提供多种自定义选项,可调整颜色、标签等。
    4. 实时更新:实时处理并可视化数据,非常适合视频分析项目。

    要生成饼图,请在上面的 Python 示例中设置 analytics_type="pie"。如需了解更多信息,请参阅指南中的可视化示例部分。

  • 可以。Analytics 解决方案内置跟踪功能:它可以实时跟踪多个对象,并每 30 帧根据跟踪对象的数据刷新图表,因此折线图、条形图、饼图和面积图都会显示实时统计数据。上面的 Python 示例中的帧循环正是这样实现的。如需了解底层跟踪功能,请参阅跟踪部分。

  • Ultralytics YOLO26 在多个方面优于 OpenCV 和 TensorFlow 等其他目标检测解决方案:

    1. 先进的准确率:YOLO26 在对象检测、实例分割、语义分割和分类任务中具有卓越的准确率。
    2. 易于使用:易用的 API 无需大量编码,即可快速实现和集成。
    3. 实时性能:经过优化,可实现高速推理,适用于实时应用。
    4. 应用广泛:支持多目标跟踪、自定义模型训练,以及导出为 ONNX、TensorRT 和 CoreML 等不同格式。
    5. 文档全面:详尽的文档和博客资源可指导用户完成每个步骤。

评论17

提出问题、分享结果,或帮助其他人。

  1. User

    Hello @glenn-jocher How we can optimize/modify the code for YOLO predict, so it can also give final compiled result of prediction as graph. In case of counting objects/cells classed as different categories in images stored in a root folder. Not just the images with the prediction boxes.

    yolo train model=yolo11n.pt data=coco8.yaml epochs=100 imgsz=640

    yolo predict model=yolo11n.pt source=path/to/bus.jpg

    Like the count graph YOLO makes after training or bar/pie chart?

    Thanks

    1. User

      Hello,

      In relation to the above question, will this code work?

      #!/usr/bin/env python3 """ yolo_predict_results.py

      Run YOLO predictions on a single image or a directory of images, save annotated images, export per-image and aggregate counts to CSV, and save plots (aggregate bar chart and optional stacked per-image bar chart).

      Requirements: pip install ultralytics pandas matplotlib pillow

      Usage (recommended - command line): # Single image python yolo_predict_results.py --model /path/to/yolo11n.pt --source /path/to/bus.jpg --output results --save-annotated --plot-aggregate

      # Directory (batch)
      python yolo_predict_results.py --model /path/to/yolo11n.pt --source /path/to/images_dir --output results --save-annotated --plot-per-image --plot-aggregate

      This file includes clearly highlighted comments showing exactly where to set your model path and images directory if you prefer to hardcode them in the script instead of using CLI args. """

      import argparse from collections import Counter from pathlib import Path import sys

      import numpy as np import pandas as pd import matplotlib.pyplot as plt from PIL import Image

      try: from ultralytics import YOLO except Exception as e: print("Please install ultralytics (pip install ultralytics).") raise

      IMAGE_EXTS = {".jpg", ".jpeg", ".png", ".bmp", ".tif", ".tiff", ".webp"}

      def find_images(directory: Path): return [str(p) for p in sorted(directory.rglob("*")) if p.suffix.lower() in IMAGE_EXTS]

      def predict_images(model, paths, conf=0.25, imgsz=640, device=None, save_annotated=False, out_dir=None): """ Run prediction per-image (keeps mapping simple). Returns list of dicts: {'file': path, 'counts': Counter, 'total': int} """ out_dir = Path(out_dir) if out_dir is not None else None annotated_dir = (out_dir / "annotated") if (save_annotated and out_dir is not None) else None if annotated_dir: annotated_dir.mkdir(parents=True, exist_ok=True)

      results = []
      for p in paths:
          print(f"Predicting: {p}")
          res_list = model.predict(source=p, conf=conf, imgsz=imgsz, device=device, verbose=False)
          if not res_list:
              print(f"Warning: no result for {p}")
              continue
          r = res_list[0]
      
          counts = Counter()
          try:
              box_cls = getattr(r.boxes, "cls", None)
              if box_cls is not None:
                  try:
                      arr = box_cls.cpu().numpy()
                  except Exception:
                      arr = np.array(box_cls)
                  arr = arr.astype(int).flatten()
                  for c in arr:
                      name = model.names.get(int(c), str(int(c)))
                      counts[name] += 1
          except Exception as e:
              print(f"Warning: could not parse classes for {p}: {e}")
      
          total = sum(counts.values())
          results.append({"file": p, "counts": counts, "total": total})
      
          if save_annotated and annotated_dir:
              try:
                  ann = r.plot()
                  if isinstance(ann, np.ndarray):
                      img_out = Image.fromarray(ann)
                  else:
                      img_out = ann
                  out_path = annotated_dir / (Path(p).stem + "_annotated" + Path(p).suffix)
                  img_out.save(out_path)
              except Exception as e:
                  print(f"Warning: failed to save annotated image for {p}: {e}")
      
      return results

      def build_dataframe(summary): all_classes = set() for row in summary: all_classes.update(row["counts"].keys()) classes = sorted(all_classes)

      rows = []
      for row in summary:
          data = {cls: row["counts"].get(cls, 0) for cls in classes}
          data["file"] = row["file"]
          data["total"] = row["total"]
          rows.append(data)
      
      if not rows:
          # Return empty dataframe
          df = pd.DataFrame(columns=["file", "total"] + classes)
      else:
          df = pd.DataFrame(rows)
          cols = ["file"] + classes + ["total"]
          df = df[cols]
      return df, classes

      def save_csvs(summary, out_dir: Path): df, classes = build_dataframe(summary) out_dir.mkdir(parents=True, exist_ok=True) csv1 = out_dir / "counts_per_image.csv" df.to_csv(csv1, index=False) print(f"Saved per-image counts to: {csv1}")

      agg = Counter()
      for r in summary:
          agg.update(r["counts"])
      agg_series = pd.Series(dict(agg)).sort_values(ascending=False)
      csv2 = out_dir / "aggregate_counts.csv"
      agg_series.to_csv(csv2, header=["count"])
      print(f"Saved aggregate counts to: {csv2}")
      return df, classes, agg

      def plot_aggregate(agg_counter, out_path: Path): names = list(agg_counter.keys()) counts = [agg_counter[n] for n in names] fig, ax = plt.subplots(figsize=(max(6, len(names) * 0.6), 5)) ax.bar(names, counts, color="tab:blue") ax.set_ylabel("Count") ax.set_title("Aggregate object counts") plt.xticks(rotation=45, ha="right") plt.tight_layout() out_path.parent.mkdir(parents=True, exist_ok=True) fig.savefig(out_path) plt.close(fig) print(f"Saved aggregate plot to: {out_path}")

      def plot_stacked_per_image(df: pd.DataFrame, classes, out_path: Path, top_n=50): if df.empty: print("No per-image data to plot.") return plot_df = df.copy() if top_n and len(plot_df) > top_n: plot_df = plot_df.sort_values("total", ascending=False).head(top_n) x = plot_df["file"].apply(lambda p: Path(p).name).tolist() bottoms = np.zeros(len(plot_df)) fig, ax = plt.subplots(figsize=(max(8, len(x) * 0.6), 6)) for cls in classes: vals = plot_df[cls].values ax.bar(x, vals, bottom=bottoms, label=cls) bottoms += vals ax.set_ylabel("Counts") ax.set_title("Per-image stacked counts") ax.legend(bbox_to_anchor=(1.05, 1), loc="upper left") plt.xticks(rotation=45, ha="right") plt.tight_layout() out_path.parent.mkdir(parents=True, exist_ok=True) fig.savefig(out_path) plt.close(fig) print(f"Saved per-image stacked plot to: {out_path}")

      def main(): ap = argparse.ArgumentParser(description="YOLO predict -> save CSV + plots") ap.add_argument("--model", required=False, help="Path to YOLO model (.pt) (e.g., yolo11n.pt)") ap.add_argument("--source", required=False, help="Image file or directory of images") ap.add_argument("--output", required=False, help="Output directory", default="results") ap.add_argument("--conf", type=float, default=0.25, help="Confidence threshold") ap.add_argument("--imgsz", type=int, default=640, help="Image size for inference") ap.add_argument("--device", type=str, default=None, help="Device (e.g., 0 or 'cpu')") ap.add_argument("--save-annotated", action="store_true", help="Save annotated images") ap.add_argument("--plot-per-image", action="store_true", help="Save stacked per-image plot") ap.add_argument("--plot-aggregate", action="store_true", help="Save aggregate bar chart") ap.add_argument("--top-n", type=int, default=50, help="Top N images for per-image plot when many images")

      # Parse CLI args (recommended usage)
      args = ap.parse_args()
      
      # -------------------------------------------------------------------------
      # HIGHLIGHT 1: If you want to hardcode paths, replace the `args = ap.parse_args()` line
      # with the block below. Edit the two highlighted lines inside the block:
      #    model="..."   <-- SET YOUR MODEL FILE PATH HERE
      #    source="..."  <-- SET YOUR IMAGE FILE OR IMAGES DIRECTORY PATH HERE
      #
      # Example (UNCOMMENT and EDIT, then COMMENT OUT or REMOVE the args = ap.parse_args() above):
      #
      # args = argparse.Namespace(
      #     model="/path/to/yolo11n.pt",       # <<< HIGHLIGHT: set model path here (absolute or relative)
      #     source="/path/to/images_or_image", # <<< HIGHLIGHT: set image file or images directory path here
      #     output="results",
      #     conf=0.25,
      #     imgsz=640,
      #     device=None,
      #     save_annotated=True,
      #     plot_per_image=True,
      #     plot_aggregate=True,
      #     top_n=50
      # )
      # -------------------------------------------------------------------------
      
      # -------------------------------------------------------------------------
      # HIGHLIGHT 2: If you prefer using the command-line, supply these two args when
      # running the script:
      #   --model /path/to/yolo11n.pt
      #   --source /path/to/images_or_image
      # These correspond to the variables used below: args.model and args.source
      # -------------------------------------------------------------------------
      
      if not args.model or not args.source:
          print("Error: --model and --source are required (or hardcode them in the script).")
          print("Example:")
          print("  python yolo_predict_results.py --model /path/to/yolo11n.pt --source /path/to/images --output results --save-annotated --plot-aggregate")
          sys.exit(1)
      
      model_path = Path(args.model)
      source = Path(args.source)
      out_dir = Path(args.output)
      out_dir.mkdir(parents=True, exist_ok=True)
      
      print(f"Loading model: {model_path}")
      model = YOLO(str(model_path))
      
      if source.is_dir():
          files = find_images(source)
          if not files:
              print(f"No images found under {source}")
              sys.exit(1)
      elif source.is_file():
          files = [str(source)]
      else:
          print(f"Source path not found: {source}")
          sys.exit(1)
      
      summary = predict_images(
          model=model,
          paths=files,
          conf=args.conf,
          imgsz=args.imgsz,
          device=args.device,
          save_annotated=args.save_annotated,
          out_dir=out_dir,
      )
      
      if not summary:
          print("No predictions were produced.")
          sys.exit(0)
      
      df, classes, agg = save_csvs(summary, out_dir)
      
      if args.plot_aggregate:
          plot_aggregate(agg, out_dir / "aggregate_counts.png")
      
      if args.plot_per_image:
          plot_stacked_per_image(df, classes, out_dir / "per_image_stacked_counts.png", top_n=args.top_n)
      
      print("All done. Results in:", out_dir.resolve())

      if name == "main": main()

      @glenn-jocher @pderrenger

    2. Glenn JocherUltralytics

      @MUK003 great question, and thanks for sharing the script in detail.

      The short answer is: yes, this is a reasonable way to get per-image and aggregate count graphs from YOLO11 predictions, since YOLO('yolo11n.pt') returns Results objects where r.boxes.cls holds the class indices and your Counter/CSV/Matplotlib logic will work on top of that. I’d recommend first running it on a very small image folder and confirming that the numbers in counts_per_image.csv and aggregate_counts.csv match what you expect from a few sample images.

      From the Ultralytics side there’s currently no built-in yolo predict CLI flag that directly produces those bar/pie charts like training; your Python approach (or, for video streams, the solutions.Analytics charts) is the right pattern. For reference on how analytics is structured inside Ultralytics, you can also glance at the Analytics solution API docs.

    1. Glenn JocherUltralytics(已编辑)

      @yeongnamtan to integrate the line graph with a line counter, use the Analytics class to update the line graph with the total object counts per frame. Ensure you initialize the Analytics with analytics_type="line" and update it within your video processing loop. For detailed guidance, please refer to the analytics section in the documentation.

    1. Glenn JocherUltralytics

      @Peymankar hello!

      To count the number of left or right turns at an intersection using Ultralytics YOLO, you can set up a tracking system that identifies and follows each vehicle's trajectory. Here’s a brief outline on how you might approach this:

      1. Track Vehicles: Use the YOLO model to detect and track vehicles. You can use the model.track() function to keep track of each vehicle's movement frame by frame.

      2. Define Zones: Define areas or lines within the video frame that represent the boundaries of different turn directions (left, right).

      3. Analyze Trajectories: Analyze the paths of tracked vehicles to determine when they cross these defined zones. This can be done by checking the intersection of vehicle tracks with these zones and classifying the movement as a left or right turn based on the zone they intersect.

      4. Count Turns: Maintain a count of how many vehicles make left or right turns based on these intersections.

      For a more detailed guide on setting up vehicle tracking and analyzing trajectories, you might find the tracking and analytics features in the Ultralytics documentation helpful. Implementing this will require some custom coding to define zones and analyze paths relative to these zones.

      If you need further assistance or more specific guidance, feel free to ask. Happy coding! 😊

  2. yeongnamtanGitHub

    ani = FuncAnimation(fig, animate, frames=range(int(cap.get(cv2.CAP_PROP_FRAME_COUNT))), blit=False, interval=1000/fps, repeat=False)

    plt.show()

    The code is still not working... I

    error meesage ... WARNING ⚠️ best.pt appears to require 'dill', which is not in ultralytics requirements. AutoInstall will run now for 'dill' but this feature will be removed in the future. Recommend fixes are to train a new model using the latest 'ultralytics' package or to run a command with an official YOLOv8 model, i.e. 'yolo predict model=yolov8n.pt' requirements: Ultralytics requirement ['dill'] not found, attempting AutoUpdate... Collecting dill Downloading dill-0.3.8-py3-none-any.whl (116 kB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 116.3/116.3 kB 4.9 MB/s eta 0:00:00 Installing collected packages: dill Successfully installed dill-0.3.8

    requirements: AutoUpdate success ✅ 5.8s, installed 1 package: ['dill'] requirements: ⚠️ Restart runtime or rerun command for updates to take effect

    requirements: Ultralytics requirement ['lapx>=0.5.2'] not found, attempting AutoUpdate... Collecting lapx>=0.5.2 Downloading lapx-0.5.9-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.7 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.7/1.7 MB 11.8 MB/s eta 0:00:00 Requirement already satisfied: numpy>=1.21.6 in /usr/local/lib/python3.10/dist-packages (from lapx>=0.5.2) (1.25.2) Installing collected packages: lapx Successfully installed lapx-0.5.9

    requirements: AutoUpdate success ✅ 7.4s, installed 1 package: ['lapx>=0.5.2'] requirements: ⚠️ Restart runtime or rerun command for updates to take effect

    0: 384x640 2 fishs, 134.7ms Speed: 13.4ms preprocess, 134.7ms inference, 747.7ms postprocess per image at shape (1, 3, 384, 640)

    DisabledFunctionError Traceback (most recent call last) in <cell line: 26>() 37 clswise_count[model.names[int(cls)]] = 1 38 ---> 39 analytics.update_bar(clswise_count) 40 cv2_imshow('Bar Plot', frame) 41 clswise_count = {}

    1 frames /usr/local/lib/python3.10/dist-packages/google/colab/_import_hooks/_cv2.py in wrapped(*args, **kwargs) 46 def wrapped(*args, **kwargs): 47 if not os.environ.get(env_var, False): ---> 48 raise DisabledFunctionError(message, name or func.name) 49 return func(*args, **kwargs) 50

    DisabledFunctionError: cv2.imshow() is disabled in Colab, because it causes Jupyter sessions to crash; see https://github.com/jupyter/notebook/issues/3935. As a substitution, consider using from google.colab.patches import cv2_imshow

    1. Paula DerrengerUltralytics

      Hello,

      Thank you for sharing the details of the issue you're encountering. It seems like the dill and lapx packages were missing but have now been successfully installed. Please ensure you restart your runtime or rerun your command to allow the updates to take effect, as suggested by the AutoUpdate messages.

      Regarding the DisabledFunctionError for cv2.imshow() in Colab, this function is indeed disabled because it can cause Jupyter sessions to crash. As an alternative, you can use cv2_imshow from the google.colab.patches module to display images in Colab notebooks. Here’s how you can modify your code:

      from google.colab.patches import cv2_imshow
      
      # Replace cv2.imshow with cv2_imshow
      cv2_imshow(frame)

      This should allow you to display images without causing any issues in your Colab environment.

      If you have any more questions or need further assistance, feel free to ask. Happy coding! 🚀

    1. Paula DerrengerUltralytics

      @yeongnamtan hello,

      It seems like you're encountering an issue with cv2.imshow not being present in the bar plot code example. This function is indeed necessary to display the frames with the visualizations in real-time. To resolve the error, you can add the following line just after updating the bar plot in the loop:

      cv2.imshow('Bar Plot', frame)

      Make sure to include this inside the loop where frames are being processed, and it will display the updated frame with the bar plot. Additionally, ensure that you have the line if cv2.waitKey(1) & 0xFF == ord('q'): to handle the window closure properly.

      If you need further assistance or have more questions, feel free to ask!

  3. yeongnamtanGitHub

    I tried to use google colab but it is giving me error because of cv2.imshow... How can I modify it to use cv2_imshow instead ... from google.colab.patches import cv2_imshow # Import cv2_imshow from google.colab.patches

    DisabledFunctionError: cv2.imshow() is disabled in Colab, because it causes Jupyter sessions to crash; see https://github.com/jupyter/notebook/issues/3935. As a substitution, consider using from google.colab.patches import cv2_imshow

    1. Paula DerrengerUltralytics

      Hello! To use cv2_imshow instead of cv2.imshow in Google Colab, you can make a simple modification to your code. First, ensure you import cv2_imshow from google.colab.patches as you've mentioned. Then, replace all instances of cv2.imshow with cv2_imshow. Here’s how you can do it:

      1. Import the function:

        from google.colab.patches import cv2_imshow
      2. Replace cv2.imshow with cv2_imshow:

        # Instead of cv2.imshow("Frame", frame)
        cv2_imshow(frame)

      This change should resolve the issue with displaying images in Google Colab. Let me know if there's anything else you need help with! 😊