-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathHOG_Kmeans.py
101 lines (95 loc) · 5.53 KB
/
HOG_Kmeans.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
"""
HOG_Kmeans.py
把5个文件的流程合在一份代码，减少了中间文件。
实现读入目标文件夹内所有图片，然后提取HOG特征，并将HOG特征作为特征向量使用kmeans算法聚类。
最后要实现每个文件夹下保存一个文件夹同名的x.csv文件，
然后还要选取数量最多的五个类，类内随机选取一张作为query
"""
import cv2
from skimage.feature import hog
import numpy as np
import pandas as pd
import os
import fnmatch
from sklearn.cluster import KMeans
import time
import joblib
debug_flag = 1
N_CLUSTERS = 9
dircount = 0
dataset_path = ("/home/zlm/dataset/vessel_reid/ALL-IMG")
for dirpath, dirs, rootfiles in os.walk(dataset_path):
for dir in dirs:
dircount+=1
savepath = os.path.join(dataset_path,dir)
if os.path.exists(savepath+"/model.pkl"):#原来的形成过，如果没改参数可以直接跳过，需要重新生成就remove
os.remove(savepath+"/model.pkl")
os.remove(savepath+"/file_label.csv")
os.remove(savepath+"/file_seleted_label.csv")
# continue
files = os.listdir(savepath)#图片列表
# print(files[0][-3:])
if not files[0][-3:]=='jpg':#如果删了前面三个文件，第一个仍不是图片，则跳过
continue
# print (dir)
start_time_hog = time.time()#计算HOG耗时
file_label_dict={}#重新清空
file_hog_dict={}
if(len(files) < N_CLUSTERS):#不够类别数目则取第一张
img_path = os.path.join(dirpath, dir, files[0])
file_seleted_label_df = pd.DataFrame({'filename':files[0],'label':0,'path':img_path},pd.Index(range(1)))
# print(file_seleted_label_df)
file_seleted_label_df.to_csv(savepath+"/file_seleted_label.csv",index=False)
outprint = "IMO:{:<10},file nember:{:<5},precent:{:<8.2%}"
outprint = outprint.format(dir, len(files), dircount/len(dirs))
continue
for filename in fnmatch.filter(files, '*.jpg'):
img_path = os.path.join(dirpath, dir, filename)
img = cv2.imread(img_path)
# Convert to grayscale and apply Gaussian filtering
# img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) #转灰度图
# im_th = cv2.adaptiveThreshold(img_gray,255,1,1,11,2) # 自适应阈值，待测试
# img_gray_cropped = img_gray[64:192, 96:288]# 裁剪坐标为[y0:y1, x0:x1]
img = cv2.resize(img, (192, 128))
# cv2.imshow('img_gray',img_gray)
# cv2.waitKey(0)
# img = np.power(img/float(np.max(img)), 1.5) #伽马矫正
hog_fd, hog_img = hog(img, orientations=9, pixels_per_cell=(8, 8), cells_per_block=(2, 2),transform_sqrt=True, visualize=True,feature_vector=True)
# print (len(hog_fd))
# cv2.imshow("hog", hog_img)
# cv2.waitKey(0)
file_label_dict.setdefault('filename',[]).append(filename)
file_label_dict.setdefault("path",[]).append(img_path)
file_hog_dict.setdefault('filename',[]).append(filename)
file_hog_dict.setdefault("path",[]).append(img_path)
hog_features = np.array(hog_fd, 'float32')
file_hog_dict.setdefault('hog',[]).append(hog_features)
end_time_hog = time.time()#计算HOG耗时
file_hog_df = pd.DataFrame(file_hog_dict)
#####################################
start_time_cluster = time.time()#计算cluster耗时
X = file_hog_df['hog'].values
X = np.vstack( X )
model=KMeans(n_clusters=N_CLUSTERS, max_iter=300)#指定分类数量，即簇的数量
label=model.fit_predict(X)#用训练器数据X拟合分类器模型并对训练器数据X进行预测
time_hog = end_time_hog - start_time_hog
time_cluster = time.time() - start_time_cluster
outprint = "IMO:{:<10},file nember:{:<5},time_hog:{:<8.3f},time_cluster:{:<8.3f},time:{:<8.3f},precent:{:<8.2%}"
outprint = outprint.format(dir, len(files), time_hog, time_cluster, time_hog+time_cluster, dircount/len(dirs))
print(outprint)
# print(dir+" file nember:"+str(len(dir))+' time_hog: %.2f time_cluster: %.2f time: %.2f' % (time_hog, time_cluster, time_hog+time_cluster)) #文件名，文件数，hog时间，cluster时间，总时间，百分比
###################################################################
joblib.dump(model, savepath + '/model.pkl')#这里要改成文件夹名。
file_label_dict["label"] = label
file_label_df = pd.DataFrame(file_label_dict)
file_label_df.to_csv(savepath+"/file_label.csv",index=False)
seleted_label = file_label_df['label'].value_counts().index.tolist()[:5]
# print(seleted_label)
file_seleted_label_df = pd.DataFrame()
for l in seleted_label:
file_seleted_label_df = file_seleted_label_df.append(file_label_df[file_label_df['label']==l].iloc[0],ignore_index=True)#从file_label_df找到label降序排序前五的一条
file_seleted_label_df['label'] = file_seleted_label_df['label'].astype(int)#因为空的df会默认用float
# print(file_seleted_label_df)
file_seleted_label_df.to_csv(savepath+"/file_seleted_label.csv",index=False)
# file_label_sorted_df = file_label_df.sort_values(by = "label")
# file_label_sorted_df.to_csv(dir+"_file_label_sorted.csv",index=False)