Technology Sharing

yolov8 训练自己的数据集

2024-07-12

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

1，准备数据集

1.1voc数据转yolo数据

voc格式

  1. 文件联级如下所述：
  2. VOCdevkit
  3.     ---VOC2007
  4.         ---Annotations
  5.         ---ImageSets
  6.         ---JPEGImages

yolo格式

  1. voc_to_yolo.py
  2.  
  3. from tqdm import tqdm
  4. import shutil
  5. from pathlib import Path
  6. import xml.etree.ElementTree as ET
  7.  
  8. def convert_label(path, lb_path, year, image_id, names):
  9.     def convert_box(size, box):
  10.         dw, dh = 1. / size[0], 1. / size[1]
  11.         x, y, w, h = (box[0] + box[1]) / 2.0 - 1, (box[2] + box[3]) / 2.0 - 1, box[1] - box[0], box[3] - box[2]
  12.         return x * dw, y * dh, w * dw, h * dh
  13.  
  14.     in_file = open(path / f'VOC{year}/Annotations/{image_id}.xml')
  15.     out_file = open(lb_path, 'w')
  16.     tree = ET.parse(in_file)
  17.     root = tree.getroot()
  18.     size = root.find('size')
  19.     w = int(size.find('width').text)
  20.     h = int(size.find('height').text)
  21.  
  22.     for obj in root.iter('object'):
  23.         cls = obj.find('name').text
  24.         if cls in names:
  25.             xmlbox = obj.find('bndbox')
  26.             bb = convert_box((w, h), [float(xmlbox.find(x).text) for x in ('xmin', 'xmax', 'ymin', 'ymax')])
  27.             cls_id = names.index(cls)  # class id
  28.             out_file.write(" ".join(str(a) for a in (cls_id, *bb)) + 'n')
  29.         else:
  30.             print("category error: ", cls)
  31.  
  32. year = "2007"
  33. image_sets = ["train", "val"]
  34. path = Path("H:\work\daodan_move\ultralytics-main\ultralytics\datasets\VOCdevkit\")
  35. class_names = ["call","dislike","fist","four","like","mute","ok","one","palm","1","2","3","4","5","6","7","8","9","10"]
  36.  
  37. for image_set in image_sets:
  38.     imgs_path = path / 'images' / f'{image_set}'
  39.     lbs_path = path / 'labels' / f'{image_set}'
  40.     imgs_path.mkdir(exist_ok=True, parents=True)
  41.     lbs_path.mkdir(exist_ok=True, parents=True)
  42.     
  43.     with open(path / f'VOC{year}/ImageSets/Main/{image_set}.txt') as f:
  44.         image_ids = f.read().strip().split()
  45.     for id in tqdm(image_ids, desc=f'{image_set}'):
  46.         f = path / f'VOC{year}/JPEGImages/{id}.jpg'  # old img path
  47.         lb_path = (lbs_path / f.name).with_suffix('.txt')  # new label path
  48.         # f.rename(imgs_path / f.name)  # move image
  49.         shutil.copyfile(f, imgs_path / f.name) # copy image
  50.         convert_label(path, lb_path, year, id, class_names)  # convert labels to YOLO format

1.2 COCO格式转YOLO格式 转换脚本

coco格式

  1. ── VOCdevkit
  2. ├── images
  3. │   ├── train	# 存放训练集图片
  4. │   └── val	# 存放验证集图片
  5. └── labels
  6.     ├── train	# 存放训练集标注文件
  7.     └── val	# 存放验证集标注文件

  1. import json
  2. import os
  3. import shutil
  4. from tqdm import tqdm
  5.  
  6. coco_path = "F:/datasets/Apple_Detection_Swift-YOLO_192"
  7. output_path = "F:/vsCode/ultralytics/datasets/Apple"
  8.  
  9. os.makedirs(os.path.join(output_path, "images", "train"), exist_ok=True)
  10. os.makedirs(os.path.join(output_path, "images", "val"), exist_ok=True)
  11. os.makedirs(os.path.join(output_path, "labels", "train"), exist_ok=True)
  12. os.makedirs(os.path.join(output_path, "labels", "val"), exist_ok=True)
  13.  
  14. with open(os.path.join(coco_path, "train", "_annotations.coco.json"), "r") as f:
  15.     train_annotations = json.load(f)
  16.  
  17. with open(os.path.join(coco_path, "valid", "_annotations.coco.json"), "r") as f:
  18.     val_annotations = json.load(f)
  19.     
  20. # Iterate over the training images
  21. for image in tqdm(train_annotations["images"]):
  22.     width, height = image["width"], image["height"]
  23.     scale_x = 1.0 / width
  24.     scale_y = 1.0 / height
  25.     
  26.     label = ""
  27.     for annotation in train_annotations["annotations"]:
  28.         if annotation["image_id"] == image["id"]:
  29.             # Convert the annotation to YOLO format
  30.             x, y, w, h = annotation["bbox"]
  31.             x_center = x + w / 2.0
  32.             y_center = y + h / 2.0
  33.             x_center *= scale_x
  34.             y_center *= scale_y
  35.             w *= scale_x
  36.             h *= scale_y
  37.             class_id = annotation["category_id"]
  38.             label += "{} {} {} {} {}n".format(class_id, x_center, y_center, w, h)
  39.             
  40.     # Save the image and label
  41.     shutil.copy(os.path.join(coco_path, "train", image["file_name"]), os.path.join(output_path, "images", "train", image["file_name"]))
  42.     with open(os.path.join(output_path, "labels", "train", image["file_name"].replace(".jpg", ".txt")), "w") as f:
  43.         f.write(label)
  44.         
  45. # Iterate over the validation images
  46. for image in tqdm(val_annotations["images"]):
  47.     width, height = image["width"], image["height"]
  48.     scale_x = 1.0 / width
  49.     scale_y = 1.0 / height
  50.     
  51.     label = ""
  52.     for annotation in val_annotations["annotations"]:
  53.         if annotation["image_id"] == image["id"]:
  54.             # Convert the annotation to YOLO format
  55.             x, y, w, h = annotation["bbox"]
  56.             x_center = x + w / 2.0
  57.             y_center = y + h / 2.0
  58.             x_center *= scale_x
  59.             y_center *= scale_y
  60.             w *= scale_x
  61.             h *= scale_y
  62.             class_id = annotation["category_id"]
  63.             label += "{} {} {} {} {}n".format(class_id, x_center, y_center, w, h)
  64.             
  65.     # Save the image and label
  66.     shutil.copy(os.path.join(coco_path, "valid", image["file_name"]), os.path.join(output_path, "images", "val", image["file_name"]))
  67.     with open(os.path.join(output_path, "labels", "val", image["file_name"].replace(".jpg", ".txt")), "w") as f:
  68.         f.write(label)

2.训练数据

将voc.yaml复制为voc_self.yaml,然后修改如下：

  1. # Ultralytics YOLO 🚀, AGPL-3.0 license
  2. # PASCAL VOC dataset http://host.robots.ox.ac.uk/pascal/VOC by University of Oxford
  3. # Documentation: # Documentation: https://docs.ultralytics.com/datasets/detect/voc/
  4. # Example usage: yolo train data=VOC.yaml
  5. # parent
  6. # ├── ultralytics
  7. # └── datasets
  8. #     └── VOC  ← downloads here (2.8 GB)
  9.  
  10. # Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
  11. path: H:\work\daodan_move\ultralytics-main\ultralytics\datasets\VOCdevkit
  12. train: # train images (relative to 'path')  16551 images
  13.   - images/train
  14. val: # val images (relative to 'path')  4952 images
  15.   - images/val
  16. test: # test images (optional)
  17.   - images/val
  18.  
  19. # Classes
  20. names:
  21.   0: call
  22.   1: dislike
  23.   2: fist
  24.   3: four
  25.   4: like
  26.   5: mute
  27.   6: ok
  28.   7: one
  29.   8: palm
  30.   9: 1
  31.   10: 2
  32.   11: 3
  33.   12: 4
  34.   13: 5
  35.   14: 6
  36.   15: 7
  37.   16: 8
  38.   17: 9
  39.   18: 10
  40.

训练数据

yolo task=detect mode=train model=yolov8n.pt data=H:\work\daodan_move\ultralytics-main\ultralytics\cfg\datasets\VOC_self.yaml epochs=100 batch=4 device=0

or

  1. # from ultralytics import YOLO
  2.  
  3. # # Load a model
  4. # model = YOLO("yolov8n.pt")  # load a pretrained model (recommended for training)
  5.  
  6. # # Train the model with 2 GPUs
  7. # results = model.train(data="H:\work\daodan_move\ultralytics-main\ultralytics\cfg\datasets\VOC_self.yaml", epochs=100, imgsz=640, device="0")
  8.  
  9. from ultralytics import YOLO
  10. if __name__ == '__main__':
  11.     #下面是三种训练的方式，使用其中一个的时候要注释掉另外两个
  12.     # Load a model  加载模型
  13.     #这种方式是选择.yaml模型文件，从零开始训练
  14.     # model = YOLO('yolov8n.yaml')  # build a new model from YAML
  15.     #这种方式是选择.pt文件，利用预训练模型训练
  16.     model = YOLO('yolov8n.pt')  # load a pretrained model (recommended for training)
  17.     #这种方式是可以加载自己想要使用的预训练模型
  18.     # model = YOLO('yolov8n.yaml').load('yolov8n.pt')  # build from YAML and transfer weights
  19.     
  20.     # Train the model   data = 这里用来设置你的data.yaml路径。即第一步设置的文件。后面的就是它的一些属性设置，你也可以增加一些比如batch=16等等。
  21.     model.train(data='H:\work\daodan_move\ultralytics-main\ultralytics\cfg\datasets\VOC_self.yaml', epochs=100, imgsz=640, device="0")

Personal profile

I have devoted myself to the research of technology for more than 30 years. I am proficient in various languages ​​such as Java, Linux, JavaScript, PHP, CSS, etc. I have made many contributions in the field of open source. I have established a developer documentation site to share some problems in technology development for everyone to read.

my contact information

Mail