一.yolov v3聚類出框
# -*- coding: utf-8 -*-
import numpy as np
import random
import argparse
import os
# # 參數名稱
# parser = argparse.ArgumentParser(description='使用該腳本生成YOLO-V3的anchor boxes\n')
# parser.add_argument('--input_annotation_txt_dir', required=True, type=str, help='輸入存儲圖片的標註txt文件(注意不要有中文)')
# parser.add_argument('--output_anchors_txt', required=True, type=str, help='輸出的存儲Anchor boxes的文本文件')
# parser.add_argument('--input_num_anchors', required=True, default=6, type=int, help='輸入要計算的聚類(Anchor boxes的個數)')
# parser.add_argument('--input_cfg_width', required=True, type=int, help="配置文件中width")
# parser.add_argument('--input_cfg_height', required=True, type=int, help="配置文件中height")
# args = parser.parse_args()
# print('args:', args)
'''
centroids 聚類點 尺寸是 numx2,類型是ndarray
annotation_array 其中之一的標註框
'''
def IOU(annotation_array, centroids):
#
similarities = []
# 其中一個標註框
w, h = annotation_array
for centroid in centroids:
c_w, c_h = centroid
if c_w >= w and c_h >= h: # 第1中情況
similarity = w * h / (c_w * c_h)
elif c_w >= w and c_h <= h: # 第2中情況
similarity = w * c_h / (w * h + (c_w - w) * c_h)
elif c_w <= w and c_h >= h: # 第3種情況
similarity = c_w * h / (w * h + (c_h - h) * c_w)
else: # 第3種情況
similarity = (c_w * c_h) / (w * h)
similarities.append(similarity)
# 將列表轉換爲ndarray
return np.array(similarities, np.float32) # 返回的是一維數組,尺寸爲(num,)
'''
k_means:k均值聚類
annotations_array 所有的標註框的寬高,N個標註框,尺寸是Nx2,類型是ndarray
centroids 聚類點 尺寸是 numx2,類型是ndarray
'''
def k_means(annotations_array, centroids, eps=0.00005, iterations=200000):
#
N = annotations_array.shape[0] # C=2
num = centroids.shape[0]
# 損失函數
distance_sum_pre = -1
assignments_pre = -1 * np.ones(N, dtype=np.int64)
#
iteration = 0
# 循環處理
while (True):
#
iteration += 1
#
distances = []
# 循環計算每一個標註框與所有的聚類點的距離(IOU)
for i in range(N):
distance = 1 - IOU(annotations_array[i], centroids)
distances.append(distance)
# 列表轉換成ndarray
distances_array = np.array(distances, np.float32) # 該ndarray的尺寸爲 Nxnum
# 找出每一個標註框到當前聚類點最近的點
assignments = np.argmin(distances_array, axis=1) # 計算每一行的最小值的位置索引
# 計算距離的總和,相當於k均值聚類的損失函數
distances_sum = np.sum(distances_array)
# 計算新的聚類點
centroid_sums = np.zeros(centroids.shape, np.float32)
for i in range(N):
centroid_sums[assignments[i]] += annotations_array[i] # 計算屬於每一聚類類別的和
for j in range(num):
centroids[j] = centroid_sums[j] / (np.sum(assignments == j))
# 前後兩次的距離變化
diff = abs(distances_sum - distance_sum_pre)
# 打印結果
print("iteration: {},distance: {}, diff: {}, avg_IOU: {}\n".format(iteration, distances_sum, diff,
np.sum(1 - distances_array) / (N * num)))
# 三種情況跳出while循環:1:循環20000次,2:eps計算平均的距離很小 3:以上的情況
if (assignments == assignments_pre).all():
print("按照前後兩次的得到的聚類結果是否相同結束循環\n")
break
if diff < eps:
print("按照eps結束循環\n")
break
if iteration > iterations:
print("按照迭代次數結束循環\n")
break
# 記錄上一次迭代
distance_sum_pre = distances_sum
assignments_pre = assignments.copy()
if __name__ == '__main__':
# 聚類點的個數,anchor boxes的個數
num_clusters = 9#args.input_num_anchors
# 索引出文件夾中的每一個標註文件的名字(.txt)
names = [i for i in os.listdir('train_images_tif_txt') if 'txt' in i]#args.input_annotation_txt_dir)
print('names:',names)
# # 標註的框的寬和高
annotations_w_h = []
for name in names:
txt_path = os.path.join('train_images_tif_txt', name)
# 讀取txt文件中的每一行
f = open(txt_path, 'r')
for line in f.readlines():
line = line.rstrip('\n')
w, h = line.split(' ')[3:] # 這時讀到的w,h是字符串類型
# eval()函數用來將字符串轉換爲數值型
annotations_w_h.append((eval(w), eval(h)))
f.close()
# 將列表annotations_w_h轉換爲numpy中的array,尺寸是(N,2),N代表多少框
annotations_array = np.array(annotations_w_h, dtype=np.float32)
N = annotations_array.shape[0]
# 對於k-means聚類,隨機初始化聚類點
random_indices = [random.randrange(N) for i in range(num_clusters)] # 產生隨機數
centroids = annotations_array[random_indices]
# k-means聚類
k_means(annotations_array, centroids, 0.00005, 200000)
# 對centroids按照寬排序,並寫入文件
widths = centroids[:, 0]
sorted_indices = np.argsort(widths)
anchors = centroids[sorted_indices]
print('anchors:',anchors)
# # 將anchor寫入文件並保存
f_anchors = open('./anchors_txt.txt', 'w')
# #
for anchor in anchors: #cfg_w train的時候用的寬度 #cfg_h train的時候用的高度
f_anchors.write('%d,%d,' % (int(anchor[0] * 200), int(anchor[1] * 1800)))
# f_anchors.write('\n')
train_images_tif_txt下存放的是如下所示的標註txt文件.
二.寬高比分析
1.kmeans.py代碼
import numpy as np
def iou(box, clusters):
"""
Calculates the Intersection over Union (IoU) between a box and k clusters.
:param box: tuple or array, shifted to the origin (i. e. width and height)
:param clusters: numpy array of shape (k, 2) where k is the number of clusters
:return: numpy array of shape (k, 0) where k is the number of clusters
"""
x = np.minimum(clusters[:, 0], box[0])
y = np.minimum(clusters[:, 1], box[1])
if np.count_nonzero(x == 0) > 0 or np.count_nonzero(y == 0) > 0:
raise ValueError("Box has no area")
intersection = x * y
box_area = box[0] * box[1]
cluster_area = clusters[:, 0] * clusters[:, 1]
iou_ = intersection / (box_area + cluster_area - intersection)
return iou_
def avg_iou(boxes, clusters):
"""
Calculates the average Intersection over Union (IoU) between a numpy array of boxes and k clusters.
:param boxes: numpy array of shape (r, 2), where r is the number of rows
:param clusters: numpy array of shape (k, 2) where k is the number of clusters
:return: average IoU as a single float
"""
return np.mean([np.max(iou(boxes[i], clusters)) for i in range(boxes.shape[0])])
def translate_boxes(boxes):
"""
Translates all the boxes to the origin.
:param boxes: numpy array of shape (r, 4)
:return: numpy array of shape (r, 2)
"""
new_boxes = boxes.copy()
for row in range(new_boxes.shape[0]):
new_boxes[row][2] = np.abs(new_boxes[row][2] - new_boxes[row][0])
new_boxes[row][3] = np.abs(new_boxes[row][3] - new_boxes[row][1])
return np.delete(new_boxes, [0, 1], axis=1)
def kmeans(boxes, k, dist=np.median):
"""
Calculates k-means clustering with the Intersection over Union (IoU) metric.
:param boxes: numpy array of shape (r, 2), where r is the number of rows
:param k: number of clusters
:param dist: distance function
:return: numpy array of shape (k, 2)
"""
rows = boxes.shape[0]
distances = np.empty((rows, k))
last_clusters = np.zeros((rows,))
np.random.seed()
print('np.random.choice(rows, k, replace=False):',np.random.choice(rows, k))
# the Forgy method will fail if the whole array contains the same rows
clusters = boxes[np.random.choice(rows, k, replace=False)]
while True:
for row in range(rows):
distances[row] = 1 - iou(boxes[row], clusters)
nearest_clusters = np.argmin(distances, axis=1)
if (last_clusters == nearest_clusters).all():
break
for cluster in range(k):
clusters[cluster] = dist(boxes[nearest_clusters == cluster], axis=0)
last_clusters = nearest_clusters
return clusters
2.example.py代碼
import glob
import xml.etree.ElementTree as ET
import cv2
import os
import numpy as np
import matplotlib.pyplot as plt
from kmeans import kmeans, avg_iou
# ANNOTATIONS_PATH = "./data/pascalvoc07-annotations"
ANNOTATIONS_PATH = "./data/widerface-annotations"
CLUSTERS = 9
# 相對原圖是否歸一化
BBOX_NORMALIZE = True
def show_cluster(data, cluster, max_points=2000):
'''
Display bouding box's size distribution and anchor generated in scatter.
'''
if len(data) > max_points:
idx = np.random.choice(len(data), max_points)
data = data[idx]
plt.scatter(data[:, 0], data[:, 1], s=5, c='lavender')
plt.scatter(cluster[:, 0], cluster[:, 1], c='red', s=100, marker="^")
plt.xlabel("Width")
plt.ylabel("Height")
plt.title("Bounding and anchor distribution")
plt.savefig("cluster.png")
plt.show()
def show_width_height(data, cluster, bins=50):
'''
Display bouding box distribution with histgram.
'''
if data.dtype != np.float32:
data = data.astype(np.float32)
width = data[:, 0]
height = data[:, 1]
ratio = height / width
plt.figure(1, figsize=(20, 6))
plt.subplot(131)
plt.hist(width, bins=bins, color='green')
plt.xlabel('width')
plt.ylabel('number')
plt.title('Distribution of Width')
plt.subplot(132)
plt.hist(height, bins=bins, color='blue')
plt.xlabel('Height')
plt.ylabel('Number')
plt.title('Distribution of Height')
plt.subplot(133)
plt.hist(ratio, bins=bins, color='magenta')
plt.xlabel('Height / Width')
plt.ylabel('number')
plt.title('Distribution of aspect ratio(Height / Width)')
plt.savefig("shape-distribution.png")
plt.show()
def sort_cluster(cluster):
'''
Sort the cluster to with area small to big.
'''
if cluster.dtype != np.float32:
cluster = cluster.astype(np.float32)
area = cluster[:, 0] * cluster[:, 1]
cluster = cluster[area.argsort()]
ratio = cluster[:, 1:2] / cluster[:, 0:1]
return np.concatenate([cluster, ratio], axis=-1)
# def load_dataset(path, normalized=True):
# '''
# load dataset from pasvoc formatl xml files
# return [[w,h],[w,h]]
# '''
# dataset = []
# for 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"))
#
# for obj in tree.iter("object"):
# if normalized:
# xmin = int(obj.findtext("bndbox/xmin")) / float(width)
# ymin = int(obj.findtext("bndbox/ymin")) / float(height)
# xmax = int(obj.findtext("bndbox/xmax")) / float(width)
# ymax = int(obj.findtext("bndbox/ymax")) / float(height)
# else:
# xmin = int(obj.findtext("bndbox/xmin"))
# ymin = int(obj.findtext("bndbox/ymin"))
# xmax = int(obj.findtext("bndbox/xmax"))
# ymax = int(obj.findtext("bndbox/ymax"))
# if (xmax - xmin) == 0 or (ymax - ymin) == 0:
# continue # to avoid divded by zero error.
# dataset.append([xmax - xmin, ymax - ymin])
#
# return np.array(dataset)
def load_dataset(path, normalized=True):
'''
load dataset from pasvoc formatl xml files
return [[w,h],[w,h]]
'''
dataset = []
names = [i for i in os.listdir(path) if 'txt' in i] # args.input_annotation_txt_dir)
# print('names:', names)
# # 標註的框的寬和高
# annotations_w_h = []
for name in names:
txt_path = os.path.join(path, name)
img_path = txt_path.replace('.txt', '.jpg')
img = cv2.imread(img_path)
img_h, img_w, _ = img.shape
# 讀取txt文件中的每一行
f = open(txt_path, 'r')
for line in f.readlines():
line = line.rstrip('\n')
w, h = line.split(' ')[3:] # 這時讀到的w,h是字符串類型
# eval()函數用來將字符串轉換爲數值型
if normalized:
dataset.append((eval(w), eval(h)))
else:
dataset.append((eval(w) * 200, eval(h) * 1800))
f.close()
return np.array(dataset)
# print("Start to load data annotations on: %s" % ANNOTATIONS_PATH)
# [[w, h], [w, h]]
data = load_dataset(path='./train_img', normalized=BBOX_NORMALIZE)
print(data[:3])
print("Start to do kmeans, please wait for a moment.")
out = kmeans(data, k=CLUSTERS)
print('==out', out)
out_sorted = sort_cluster(out)
print("Accuracy: {:.2f}%".format(avg_iou(data, out) * 100))
#
show_cluster(data, out, max_points=2000)
if out.dtype != np.float32:
out = out.astype(np.float32)
print("Recommanded aspect ratios(width/height)")
print("Width Height Height/Width")
for i in range(len(out_sorted)):
print("%.3f %.3f %.1f" % (out_sorted[i, 0], out_sorted[i, 1], out_sorted[i, 2]))
show_width_height(data, out, bins=50)
txt是類別, cx,cy,w和h是歸一化後的比例),下圖是其分佈,也就是輸入如果是方形,anchor ratio比就用這個
下圖是乘以實際尺寸後的分佈,也就是輸入如果是圖片等比例 anchor ratio比就用這個