kmeans以及kmeans++聚类生成anchors

使用yolo系列通常需要通过聚类算法生成anchors，本文给出kmeans以及kmeans++的python实现。
数据格式为VOC的xml文件

若数据集不是voc格式，比如coco格式或者txt格式，可以采用如下方法进行数据集格式转换转为voc再进行聚类计算先验框：
https://blog.csdn.net/Arcofcosmos/article/details/119983270?spm=1001.2014.3001.5502

kmeans聚类

'''
Author: TuZhou
Version: 1.0
Date: 2021-08-22 18:39:19
LastEditTime: 2021-09-12 17:14:13
LastEditors: TuZhou
Description:
FilePath: \My_Mobile_Yolo2\utils\kmeans_for_anchors.py
'''
#-------------------------------------------------------------------------------------------------#
#   kmeans虽然会对数据集中的框进行聚类，但是很多数据集由于框的大小相近，聚类出来的9个框相差不大，
#   这样的框反而不利于模型的训练。因为不同的特征层适合不同大小的先验框，越浅的特征层适合越大的先验框
#   原始网络的先验框已经按大中小比例分配好了，不进行聚类也会有非常好的效果。
#-------------------------------------------------------------------------------------------------#
import glob
import xml.etree.ElementTree as ETimport numpy as npdef cas_iou(box,cluster):x = np.minimum(cluster[:,0],box[0])y = np.minimum(cluster[:,1],box[1])intersection = x * yarea1 = box[0] * box[1]area2 = cluster[:,0] * cluster[:,1]iou = intersection / (area1 + area2 -intersection)return ioudef avg_iou(box,cluster):return np.mean([np.max(cas_iou(box[i],cluster)) for i in range(box.shape[0])])def kmeans(box,k):# 取出一共有多少框row = box.shape[0]# 每个框各个点的位置distance = np.empty((row,k))# 最后的聚类位置last_clu = np.zeros((row,))np.random.seed()# 随机选5个当聚类中心cluster = box[np.random.choice(row,k,replace = False)]# cluster = random.sample(row, k)while True:# 计算每一行距离五个点的iou情况。for i in range(row):distance[i] = 1 - cas_iou(box[i],cluster)# 取出最小点near = np.argmin(distance,axis=1)if (last_clu == near).all():break# 求每一个类的中位点for j in range(k):cluster[j] = np.median(box[near == j],axis=0)last_clu = nearreturn clusterdef load_data(path):data = []# 对于每一个xml都寻找boxfor xml_file in glob.glob('{}/*xml'.format(path)):tree = ET.parse(xml_file)height = int(tree.findtext('./size/height'))width = int(tree.findtext('./size/width'))if height<=0 or width<=0:continue# 对于每一个目标都获得它的宽高for obj in tree.iter('object'):xmin = int(float(obj.findtext('bndbox/xmin'))) / widthymin = int(float(obj.findtext('bndbox/ymin'))) / heightxmax = int(float(obj.findtext('bndbox/xmax'))) / widthymax = int(float(obj.findtext('bndbox/ymax'))) / heightxmin = np.float64(xmin)ymin = np.float64(ymin)xmax = np.float64(xmax)ymax = np.float64(ymax)# 得到宽高data.append([xmax-xmin,ymax-ymin])return np.array(data)def calculate_anchors(dataset_anno_path = r'./datasets/RoadSignsPascalVOC/Annotations', anchorsPath = './datasets/RoadSignsPascalVOC/yolo_anchors.txt', anchors_num = 9, SIZE = 416):# 运行该程序会计算'./VOCdevkit/VOC2007/Annotations'的xml# 会生成yolo_anchors.txt# SIZE = 416# anchors_num = 9# # 载入数据集，可以使用VOC的xml# #path = r'./VOCdevkit/VOC2007/Annotations'# path = r'./datasets/WZRY/Annotations'# 载入所有的xml# 存储格式为转化为比例后的width,heightdata = load_data(dataset_anno_path)# 使用k聚类算法out = kmeans(data,anchors_num)out = out[np.argsort(out[:,0])]print('acc:{:.2f}%'.format(avg_iou(data,out) * 100))print(out*SIZE)data = out*SIZEf = open(anchorsPath, 'w')row = np.shape(data)[0]for i in range(row):if i == 0:x_y = "%d,%d" % (data[i][0], data[i][1])else:x_y = ", %d,%d" % (data[i][0], data[i][1])f.write(x_y)f.close()if __name__ == '__main__':#数据集xml注释路径dataset_anno_path = r'./datasets/RoadSignsPascalVOC/Annotations'#生成的anchors的txt文件保存路径anchorsPath = './datasets/RoadSignsPascalVOC/yolo_anchors.txt'#生成的anchors数量anchors_num = 9#输入的图片尺寸SIZE = 416calculate_anchors(dataset_anno_path, anchorsPath, anchors_num, SIZE)

kmeans++聚类

'''
Author: TuZhou
Version: 1.0
Date: 2021-09-12 15:53:55
LastEditTime: 2021-09-12 17:14:55
LastEditors: TuZhou
Description:
FilePath: \My_Mobile_Yolo2\utils\k_means_plus.py
'''import glob
import xml.etree.ElementTree as ETimport numpy as npdef cas_iou(box,cluster):x = np.minimum(cluster[:,0],box[0])y = np.minimum(cluster[:,1],box[1])intersection = x * yarea1 = box[0] * box[1]area2 = cluster[:,0] * cluster[:,1]iou = intersection / (area1 + area2 -intersection)return ioudef avg_iou(box,cluster):return np.mean([np.max(cas_iou(box[i],cluster)) for i in range(box.shape[0])])def bboxesOverRation(bboxesA,bboxesB):"""功能等同于matlab的函数bboxesOverRationbboxesA：M*4 array,形如[x,y,w,h]排布bboxesB: N*4 array,形如[x,y,w,h]排布"""bboxesA = np.array(bboxesA.astype('float'))bboxesB = np.array(bboxesB.astype('float'))M = bboxesA.shape[0]N = bboxesB.shape[0]areasA = bboxesA[:,2]*bboxesA[:,3]areasB = bboxesB[:,2]*bboxesB[:,3]xA = bboxesA[:,0]+bboxesA[:,2]yA = bboxesA[:,1]+bboxesA[:,3]xyA = np.stack([xA,yA]).transpose()xyxyA = np.concatenate((bboxesA[:,:2],xyA),axis=1)xB = bboxesB[:,0] +bboxesB[:,2]yB = bboxesB[:,1]+bboxesB[:,3]xyB = np.stack([xB,yB]).transpose()xyxyB = np.concatenate((bboxesB[:,:2],xyB),axis=1)iouRatio = np.zeros((M,N))for i in range(M):for j in range(N):x1 = max(xyxyA[i,0],xyxyB[j,0]);x2 = min(xyxyA[i,2],xyxyB[j,2]);y1 = max(xyxyA[i,1],xyxyB[j,1]);y2 = min(xyxyA[i,3],xyxyB[j,3]);Intersection = max(0,(x2-x1))*max(0,(y2-y1));Union = areasA[i]+areasB[j]-Intersection;iouRatio[i,j] = Intersection/Union; return iouRatiodef load_data(path):data = []# 对于每一个xml都寻找boxfor xml_file in glob.glob('{}/*xml'.format(path)):tree = ET.parse(xml_file)height = int(tree.findtext('./size/height'))width = int(tree.findtext('./size/width'))if height<=0 or width<=0:continue# 对于每一个目标都获得它的宽高for obj in tree.iter('object'):xmin = int(float(obj.findtext('bndbox/xmin'))) / widthymin = int(float(obj.findtext('bndbox/ymin'))) / heightxmax = int(float(obj.findtext('bndbox/xmax'))) / widthymax = int(float(obj.findtext('bndbox/ymax'))) / heightxmin = np.float64(xmin)ymin = np.float64(ymin)xmax = np.float64(xmax)ymax = np.float64(ymax)# 得到宽高x = xmin + 0.5 * (xmax-xmin)y = ymin + 0.5 * (ymax-ymin)data.append([x, y, xmax-xmin,ymax-ymin])return np.array(data)def estimateAnchorBoxes(trainingData,numAnchors=9,SIZE = 416):'''功能：kmeans++算法估计anchor，类似于matlab函数estimateAnchorBoxes,当trainingData数据量较大时候，自写的kmeans迭代循环效率较低，matlab的estimateAnchorBoxes得出anchors较快，但meanIOU较低，然后乘以实际box的ratio即可。此算法由于优化是局部，易陷入局部最优解，结果不一致属正常cuixingxing150@gmail.comExample: import scipy.io as scipodata = scipo.loadmat(r'D:\Matlab_files\trainingData.mat')trainingData = data['temp']meanIoUList = []for numAnchor in np.arange(1,16):anchorBoxes,meanIoU = estimateAnchorBoxes(trainingData,numAnchors=numAnchor)meanIoUList.append(meanIoU)plt.plot(np.arange(1,16),meanIoUList,'ro-')plt.ylabel("Mean IoU")plt.xlabel("Number of Anchors")plt.title("Number of Anchors vs. Mean IoU")Parameters----------trainingData : numpy 类型形如[x,y,w,h]排布，M*4大小二维矩阵numAnchors : int, optional估计的anchors数量. The default is 9.Returns-------anchorBoxes : numpy类型形如[w,h]排布，N*2大小矩阵.meanIoU : scalar 标量DESCRIPTION.'''numsObver = trainingData.shape[0]xyArray = np.zeros((numsObver,2))trainingData[:,0:2] = xyArrayassert(numsObver>=numAnchors)# kmeans++# init centroids = [] # 初始化中心，kmeans++centroid_index = np.random.choice(numsObver, 1)centroids.append(trainingData[centroid_index])while len(centroids)<numAnchors:minDistList = []for box in trainingData:box = box.reshape((-1,4))minDist = 1for centroid in centroids:centroid = centroid.reshape((-1,4))ratio = (1-bboxesOverRation(box,centroid)).item()if ratio<minDist:minDist = ratiominDistList.append(minDist)sumDist = np.sum(minDistList)prob = minDistList/sumDist idx = np.random.choice(numsObver,1,replace=True,p=prob)centroids.append(trainingData[idx])# kmeans 迭代聚类maxIterTimes = 100iter_times = 0while True:minDistList = []minDistList_ind = []for box in trainingData:box = box.reshape((-1,4))minDist = 1box_belong = 0for i,centroid in enumerate(centroids):centroid = centroid.reshape((-1,4))ratio = (1-bboxesOverRation(box,centroid)).item()if ratio<minDist:minDist = ratiobox_belong = iminDistList.append(minDist)minDistList_ind.append(box_belong)centroids_avg = []for _ in range(numAnchors):centroids_avg.append([])for i,anchor_id in enumerate(minDistList_ind):centroids_avg[anchor_id].append(trainingData[i])err = 0for i in range(numAnchors):if len(centroids_avg[i]):temp = np.mean(centroids_avg[i],axis=0)err +=  np.sqrt(np.sum(np.power(temp-centroids[i],2)))centroids[i] = np.mean(centroids_avg[i],axis=0)iter_times+=1if iter_times>maxIterTimes or err==0:breakanchorBoxes = np.array([x[2:] for x in centroids])meanIoU = 1-np.mean(minDistList)anchorBoxes = anchorBoxes[np.argsort(anchorBoxes[:,0])]print('acc:{:.2f}%'.format(avg_iou(trainingData[:,2:],anchorBoxes) * 100))anchorBoxes = anchorBoxes*SIZEreturn anchorBoxes,meanIoUdef calculate_anchors(dataset_anno_path = r'./datasets/RoadSignsPascalVOC/Annotations', anchorsPath = './yolo_anchors.txt', anchors_num = 9, SIZE = 416):data = load_data(dataset_anno_path)anchors, _ = estimateAnchorBoxes(data, numAnchors = anchors_num, SIZE = SIZE)print(anchors)f = open(anchorsPath, 'w')row = np.shape(anchors)[0]for i in range(row):if i == 0:x_y = "%d,%d" % (anchors[i][0], anchors[i][1])else:x_y = ", %d,%d" % (anchors[i][0], anchors[i][1])f.write(x_y)f.close()if __name__ == "__main__":#数据集xml注释路径dataset_anno_path = r'./datasets/RoadSignsPascalVOC/Annotations'#生成的anchors的txt文件保存路径anchorsPath = './datasets/RoadSignsPascalVOC/yolo_anchors.txt'#生成的anchors数量anchors_num = 9#输入的图片尺寸SIZE = 416calculate_anchors(dataset_anno_path, anchorsPath, anchors_num, SIZE)

参考自

kmeans以及kmeans++聚类生成anchors相关推荐

kmeans++聚类生成anchors
kmeans++聚类生成anchors 说明使用yolo系列通常需要通过kmeans聚类算法生成anchors, 但kmeans算法本身具有一定的局限性,聚类结果容易受初始值选取影响. 因此通过改进 ...
R语言生成仿真的3D高斯簇数据集、使用scale函数进行数据缩放、并使用KMeans进行聚类分析、数据反向缩放并比较聚类生成的中心和实际数据的中心的差异、预测新的数据所属的聚类簇
R语言生成仿真的3D高斯簇数据集.使用scale函数进行数据缩放.并使用KMeans进行聚类分析.数据反向缩放并比较聚类生成的中心和实际数据的中心的差异.预测新的数据所属的聚类簇目录
kmeans算法实现及获取anchors
kmeans算法网上资料很多,其原理简单来说就选取k个聚类中心,然后把剩余的点归类到与其相距最近的聚类中心,是一种无监督算法.缺点和不足有两个,第一个是k需要指定,第二个是对于聚类中心种子点的选取很敏 ...
C#下实现的基础K-MEANS多维聚类
转自:http://www.cnblogs.com/lzxwalex/p/7708266.html C#下实现的基础K-MEANS多维聚类 #本文PDF版下载 C#下实现的基础K-MEANS多维聚类P ...
聚类分析 | MATLAB实现k-Means(k均值聚类)分析
目录聚类分析 | MATLAB实现k-Means(k均值聚类)分析 k-均值聚类简介相关描述程序设计学习小结参考资料致谢聚类分析 | MATLAB实现k-Means(k均值聚类)分析 k ...
kmeans算法_KMeans聚类算法详解
1. 写在前面如果想从事数据挖掘或者机器学习的工作,掌握常用的机器学习算法是非常有必要的,常见的机器学习算法: 监督学习算法:逻辑回归,线性回归,决策树,朴素贝叶斯,K近邻,支持向量机,集成算法Ad ...
关于距离，K-means，层次聚类，密度聚类以及谱聚类
目录相似度/距离 k-Means算法衡量聚类(轮廓系数) 层次聚类密度聚类谱聚类之前博客中讲的模型基本上都是分类以及回归模型,他们都是属于有监督学习的,意为所有的样本都有一个结果值提供,我们 ...
机器学习 K-Means 实现文本聚类 2021-10-30
人工智能总目录新闻头条数据进行聚类分析人工智能总目录 1. 数据集信息 2. 数据预处理 2.1 为向量化表示进行前处理 2.2 TF-IDF 2.3 Stemming 2.4 Tokenizin ...
K-means和PAM聚类算法Python实现及对比
K-means(K均值划分)聚类:简单的说,一般流程如下:先随机选取k个点,将每个点分配给它们,得到最初的k个分类:在每个分类中计算均值,将点重新分配,划归到最近的中心点:重复上述步骤直到点的划归不再 ...

kmeans以及kmeans++聚类生成anchors

kmeans以及kmeans++聚类生成anchors相关推荐

最新文章

热门文章