数据集格式

训练代码

from ultralytics import YOLO
import multiprocessing

if __name__ == '__main__':
    # 添加 Windows 多进程支持
    multiprocessing.freeze_support()
    # Load a model
    model = YOLO("yolo11n.pt")  # load a pretrained model (recommended for training)

    # Train the model with MPS
    results = model.train(data="coco8.yaml", epochs=100, imgsz=640)
    # 量化
    results = model.val(half=True)  # evaluate model performance on the validation set
    # Run inference with the model
    # results = model("datasets/coco8/images/val/000000000061.jpg")  # predict on an image
    # result.save(filename="result.jpg")
    # 另存为模型
    model.save("best.pt")

验证识别

from ultralytics import YOLO
# Load a model
model = YOLO("best.pt")  # pretrained YOLO11n model

# Run batched inference on a list of images
results = model(["datasets/coco8/images/train/000000000034.jpg"])  # return a list of Results objects

# Process results list
for result in results:
    boxes = result.boxes  # Boxes object for bounding box outputs
    masks = result.masks  # Masks object for segmentation masks outputs
    keypoints = result.keypoints  # Keypoints object for pose outputs
    probs = result.probs  # Probs object for classification outputs
    obb = result.obb  # Oriented boxes object for OBB outputs
    result.show()  # display to screen
    result.save(filename="result.jpg")  # save to disk

基于模型做成一个在线api接口

import torch
from flask import Flask, request, jsonify
from ultralytics import YOLO
import cv2
import numpy as np
import base64
from PIL import Image
import io

app = Flask(__name__)

# 检查 CUDA 是否可用
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")

model = YOLO("best.pt").to(device)

def process_image(image_data):
    results = model(image_data)
    result = results[0]
    
    # 获取检测结果
    detections = []
    for box in result.boxes:
        detection = {
            "bbox": box.xyxy[0].tolist(),  # 边界框坐标
            "confidence": float(box.conf),  # 置信度
            "class_id": int(box.cls),      # 类别ID
        }
        detections.append(detection)
    
    # 保存结果图像
    result_img = result.plot()
    
    # 将结果图像转换为base64
    is_success, buffer = cv2.imencode(".jpg", result_img)
    if is_success:
        img_base64 = base64.b64encode(buffer).decode('utf-8')
        return detections, img_base64
    return None, None

@app.route('/predict', methods=['POST'])
def predict():
    try:
        # 检查是否有文件上传
        if 'image' not in request.files:
            return jsonify({'error': 'No image uploaded'}), 400
        
        file = request.files['image']
        # 读取图像文件
        image_bytes = file.read()
        image = Image.open(io.BytesIO(image_bytes))
        
        # 处理图像
        detections, img_base64 = process_image(image)
        
        if detections is None:
            return jsonify({'error': 'Failed to process image'}), 500
        
        # 返回结果
        return jsonify({
            'detections': detections,
            'image': img_base64
        })
        
    except Exception as e:
        return jsonify({'error': str(e)}), 500

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5000, debug=True)