def facescrub_db():
'''
dbscan train time: 508 s
core cluster num: 92144
gt n_cluster: 530
pre n_cluster: 509
evaluate score: {'ARI': 0.7734228809345598, 'AMI': 0.9438636008104975, 'HCV': (0.9544715446091917, 0.9783882920765555, 0.9662819482731974), 'FMI': 0.7872490272398316}
unsupervised evaluate score: {'SC': 0.5033557155857413, 'CH': 292.44290237535967, 'DBI': 1.9704205586329755}
:return:
'''
feature_path = 'C:\\Users\\xiongraorao\\Desktop\\facescrub_feature.txt'
X, y, img_path = load_feature(feature_path)
dbscan = DBSCAN(n_jobs=3, eps=0.5, min_samples=10, metric='cosine')
start = time.time()
dbscan.fit(X)
print('dbscan train time: %d s' % round(time.time() - start))
print('core cluster num: ', len(dbscan.core_sample_indices_))
print('gt n_cluster: ', len(set(y)))
print('pre n_cluster: ', len(set(dbscan.labels_)))
# evaluate
score = cluster_eval(y, dbscan.labels_)
print('evaluate score: ', score)
unsupervise_score = cluster_eval_withoutgt(X, dbscan.labels_)
print('unsupervised evaluate score: ', unsupervise_score)
return img_path, dbscan.labels_
def lfw_kmeans():
'''
result:
817s
gt n_cluster: 5740
pre n_cluster: 5740
evaluate score: {'ARI': 0.2946529011301506, 'AMI': 0.7480503460467403, 'HCV': (0.9914706899544379, 0.9471878835541047, 0.9688235331962477), 'FMI': 0.41122064384553325}
unsupervised evaluate score: 0.2398293099762878
:return:
'''
path = 'C:\\Users\\xiongraorao\\Desktop\\lfw_feature.txt'
X, gt_label, img_path = load_feature(path)
kmeans = KMeans(n_jobs=-1, n_clusters=5740)
start = time.time()
kmeans.fit(X)
print('gt n_cluster: ', len(set(gt_label)))
print('pre n_cluster: ', len(set(kmeans.labels_)))
print('cost time: %d s' % round(time.time() - start))
# # evaluate
#
# score = cluster_eval(gt_label, kmeans.labels_)
# print('evaluate score: ', score)
#
# unsupervise_score = cluster_eval_withoutgt(X, kmeans.labels_)
# print('unsupervised evaluate score: ', unsupervise_score)
return img_path, kmeans.labels_
def evaluate_db():
'''
从数据库中得到pre_cluster
:param gt:
:return:
'''
feature_path = 'C:\\Users\\xiongraorao\\Desktop\\facescrub_feature.txt'
_, y, img_path = load_feature(feature_path)
pre_label = []
db = Mysql('192.168.1.11', 3306, 'root', '123456', 'face_reid')
logger = Log('visualize', is_save=False)
db.set_logger(logger)
from tqdm import tqdm
for img in tqdm(img_path):
file_name = img.split('/')[-1]
sql = 'select cluster_id from t_cluster2 where uri like {}'.format(
"\"%" + file_name + "\"")
label = db.select(sql)
pre_label.append(label[0])
# evaluate
assert len(y) == len(
pre_label), 'gt and predict label have different length'
score = cluster_eval(y, pre_label)
print('evaluate score: ', score)
def lfw_AP():
'''
gt n_cluster: 5740
pre n_cluster: 1723
cost time: 275 s
evaluate score: {'ARI': 0.46572507910154803, 'AMI': 0.6476644589883233, 'HCV': (0.8942740529630575, 0.9747204563661763, 0.9327659430454032), 'FMI': 0.506312152565195}
unsupervised evaluate score: {'SC': 0.24097617806939176, 'CH': 7.922003068266033, 'DBI': 2.010803466033437}
:return:
'''
path = 'C:\\Users\\xiongraorao\\Desktop\\lfw_feature.txt'
X, gt_label, img_path = load_feature(path)
ap = AffinityPropagation(affinity='euclidean')
start = time.time()
ap.fit(X)
print('gt n_cluster: ', len(set(gt_label)))
print('pre n_cluster: ', len(set(ap.labels_)))
print('cost time: %d s' % round(time.time() - start))
# evaluate
score = cluster_eval(gt_label, ap.labels_)
print('evaluate score: ', score)
unsupervise_score = cluster_eval_withoutgt(X, ap.labels_)
print('unsupervised evaluate score: ', unsupervise_score)
return img_path, ap.labels_
def lfw_birch():
'''
gt n_cluster: 5740
pre n_cluster: 5740
cost time: 16 s
evaluate score: {'ARI': 0.9088314702653802, 'AMI': 0.9457411154577685, 'HCV': (0.9950424072953659, 0.9883225758324755, 0.9916711078460287), 'FMI': 0.911949218026558}
unsupervised evaluate score: 0.34134643599754744
:return:
'''
path = 'C:\\Users\\xiongraorao\\Desktop\\lfw_feature.txt'
X, gt_label, img_path = load_feature(path)
birch = Birch(n_clusters=5740)
start = time.time()
birch.fit(X)
print('gt n_cluster: ', len(set(gt_label)))
print('pre n_cluster: ', len(set(birch.labels_)))
print('cost time: %d s' % round(time.time() - start))
# evaluate
score = cluster_eval(gt_label, birch.labels_)
print('evaluate score: ', score)
unsupervise_score = cluster_eval_withoutgt(X, birch.labels_)
print('unsupervised evaluate score: ', unsupervise_score)
return img_path, birch.labels_
def lfw2():
path = 'C:\\Users\\xiongraorao\\Desktop\\lfw_feature.txt'
X, gt_label, img_path = load_feature(path)
# 分成5份依次聚类
xx = split_list(X, 5)
for i in range(5):
db = DBSCAN(n_jobs=-1, eps=0.5, min_samples=1, metric='cosine')
def insert_ground_truth():
'''
写入真实样本的标签
:param files_path:
:param gt_labels:
:return:
'''
# feature_path = 'C:\\Users\\xiongraorao\\Desktop\\facescrub_feature.txt'
feature_path = 'C:\\Users\\xiongraorao\\Desktop\\lfw_feature.txt'
_, y, img_path = load_feature(feature_path)
return img_path, y
def lfw_agglomerative(linkage='complete'):
'''
# complete cosine
gt n_cluster: 5740
pre n_cluster: 5740
cost time: 34 s
evaluate score: {'ARI': 0.8772158012260838, 'AMI': 0.9540866194718837, 'HCV': (0.9970073581889802, 0.9901151744759075, 0.993549313824344), 'FMI': 0.8836290092881737}
unsupervised evaluate score: {'SC': 0.3660289160550356, 'CH': 6.788387059880234, 'DBI': 0.6420190135471637}
# average cosine
gt n_cluster: 5740
pre n_cluster: 5740
cost time: 33 s
evaluate score: {'ARI': 0.9902014517124091, 'AMI': 0.985990567106264, 'HCV': (0.9975087289538368, 0.9969732346023019, 0.9972409098911819), 'FMI': 0.990228621069258}
unsupervised evaluate score: {'SC': 0.40472477697699766, 'CH': 6.732017848985664, 'DBI': 0.6379048022339981}
# single cosine
gt n_cluster: 5740
pre n_cluster: 5740
cost time: 30 s
evaluate score: {'ARI': 0.9882399651579442, 'AMI': 0.9857927097057834, 'HCV': (0.9969255699362045, 0.9971855683313122, 0.9970555521840598), 'FMI': 0.9882756231149297}
unsupervised evaluate score: {'SC': 0.40295206143306594, 'CH': 6.662192377284881, 'DBI': 0.7650319783317995}
:return:
'''
path = 'C:\\Users\\xiongraorao\\Desktop\\lfw_feature.txt'
X, gt_label, img_path = load_feature(path)
agg = AgglomerativeClustering(n_clusters=5740,
affinity='cosine',
linkage=linkage)
start = time.time()
agg.fit(X)
print('gt n_cluster: ', len(set(gt_label)))
print('pre n_cluster: ', len(set(agg.labels_)))
print('cost time: %d s' % round(time.time() - start))
# evaluate
score = cluster_eval(gt_label, agg.labels_)
print('evaluate score: ', score)
unsupervise_score = cluster_eval_withoutgt(X, agg.labels_)
print('unsupervised evaluate score: ', unsupervise_score)
return img_path, agg.labels_
def lfw():
'''
result:
dbscan train time: 9 s
core cluster num: 13211
gt n_cluster: 5740
pre n_cluster: 5768
evaluate score: {'ARI': 0.9884690711797678, 'AMI': 0.9851659125925958, 'HCV': (0.9973329806102722, 0.9967953906459855, 0.9970641131646472), 'FMI': 0.9885028490650399}
unsupervised evaluate score: {'SC': 0.40285973649881957, 'CH': 6.693587946673831, 'DBI': 0.6624322266677443}
:return:
'''
path = 'C:\\Users\\xiongraorao\\Desktop\\lfw_feature.txt'
X, gt_label, img_path = load_feature(path)
## DBSCAN cluster
dbscan = DBSCAN(n_jobs=-1, eps=0.5, min_samples=1, metric='cosine')
start = time.time()
dbscan.fit(X)
print('dbscan train time: %d s' % round(time.time() - start))
print('core cluster num: ', len(dbscan.core_sample_indices_))
print('gt n_cluster: ', len(set(gt_label)))
print('pre n_cluster: ', len(set(dbscan.labels_)))
# evaluate
score = cluster_eval(gt_label, dbscan.labels_)
print('evaluate score: ', score)
unsupervise_score = cluster_eval_withoutgt(X, dbscan.labels_)
print('unsupervised evaluate score: ', unsupervise_score)
return img_path, dbscan.labels_