yolov5里的K-means代码改为K-means++

Yolov5中的K-means代码实现了经典的K-means聚类算法，如果要改为K-means++，可以按照以下步骤进行修改：

1. 修改kmeans.py文件中的kmeans函数，将初始化质心的方式改为K-means++算法。具体来说，可以参考以下代码：

def kmeans_plus_plus(X, n_clusters, random_state=0):
    rng = np.random.RandomState(random_state)
    n_samples = X.shape[0]
    centers = [X[rng.randint(n_samples)]]
    distances = np.zeros(n_samples)
    
    for i in range(1, n_clusters):
        for j in range(n_samples):
            distances[j] = np.min([np.linalg.norm(X[j] - c) for c in centers])
        probs = distances / np.sum(distances)
        cumprobs = np.cumsum(probs)
        r = rng.rand()
        for j, cp in enumerate(cumprobs):
            if cp > r:
                break
        centers.append(X[j])
    return centers

def kmeans(X, n_clusters, random_state=0):
    centers = kmeans_plus_plus(X, n_clusters, random_state)
    labels = np.zeros(X.shape[0])
    while True:
        old_centers = np.copy(centers)
        for i in range(X.shape[0]):
            distances = np.linalg.norm(X[i] - centers, axis=1)
            labels[i] = np.argmin(distances)
        for j in range(n_clusters):
            centers[j] = np.mean(X[labels == j], axis=0)
        if np.allclose(old_centers, centers):
            break
    return labels

2. 在train.py文件中，找到以下代码行：

anchors = kmeans(wh, num_anchors, random_state=0)

将其修改为以下代码：

anchors = kmeans_plus_plus(wh, num_anchors, random_state=0)

这样就可以将Yolov5中的K-means聚类算法改为K-means++算法。