【光电智造】微调YOLOv11：实用指南【详细过程与代码】

import jsonimport os
def convert_coco_to_yolo(coco_file, output_dir):    with open(coco_file) as f:        data = json.load(f)        for image in data['images']:        annotations = [ann for ann in data['annotations'] if ann['image_id'] == image['id']]        label_file = os.path.join(output_dir, f"{image['file_name'].split('.')[0]}.txt")        with open(label_file, 'w') as f:            for ann in annotations:                category_id = ann['category_id'] - 1  # YOLO classes are 0-indexed                bbox = ann['bbox']                x_center = (bbox[0] + bbox[2] / 2) / image['width']                y_center = (bbox[1] + bbox[3] / 2) / image['height']                width = bbox[2] / image['width']                height = bbox[3] / image['height']                f.write(f"{category_id} {x_center} {y_center} {width} {height}\n")

import albumentations as Afrom albumentations.pytorch import ToTensorV2
transform = A.Compose([    A.RandomCrop(width=640, height=640),    A.HorizontalFlip(p=0.5),    A.RandomBrightnessContrast(p=0.2),    A.HueSaturationValue(p=0.2),    ToTensorV2()])

from sklearn.model_selection import train_test_splitimport osimport shutil
def split_dataset(images_dir, labels_dir, output_dir, test_size=0.2, val_size=0.2):    images = [f for f in os.listdir(images_dir) if f.endswith('.jpg')]    train_images, test_images = train_test_split(images, test_size=test_size, random_state=42)    train_images, val_images = train_test_split(train_images, test_size=val_size, random_state=42)
    for subset, subset_images in [('train', train_images), ('val', val_images), ('test', test_images)]:        os.makedirs(f"{output_dir}/images/{subset}", exist_ok=True)        os.makedirs(f"{output_dir}/labels/{subset}", exist_ok=True)        for image in subset_images:            shutil.copy(f"{images_dir}/{image}", f"{output_dir}/images/{subset}/{image}")            label_file = image.replace('.jpg', '.txt')            shutil.copy(f"{labels_dir}/{label_file}", f"{output_dir}/labels/{subset}/{label_file}")

path: ../datasets  # Path to dataset root directorytrain: images/train  # Path to training imagesval: images/val  # Path to validation imagesnc: 3  # Number of classesnames: ['class1', 'class2', 'class3']  # Class names

from ultralytics import YOLO
model = YOLO('yolov8n.pt')  # Load YOLOv8 Nano pretrained weights

model.train(data='custom_dataset.yaml',  # Path to YAML config            epochs=50,                  # Number of epochs            imgsz=640,                  # Image size            batch=16,                   # Batch size            device=0)                   # GPU device index

model.train(data='custom_dataset.yaml',            epochs=50,            imgsz=640,            lr0=0.01,  # Starting learning rate            optimizer='AdamW',            augment=True)