-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathclustering.py
132 lines (122 loc) · 6.32 KB
/
clustering.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
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
import os
import cv2
import numpy as np
from PIL import Image
from sklearn.cluster import MiniBatchKMeans
from collections import Counter
import datetime
import argparse
import json
def main(conf, clustering_algorithm):
model_folder = config['trainer']['save_dir'] + config['experim_name']
### VOC Dataset
if config['dataset'] == 'voc':
label_folder = 'datasets/voc/VOCdevkit/VOC2012/SegmentationClassAug'
if conf['n_labeled_examples'] == 662:
split_list = [132, 2, 1, 1, 1, 2, 3, 4, 7, 2, 1, 2, 6, 2, 2, 15, 1, 1, 2, 2, 1]
elif conf['n_labeled_examples'] == 331:
split_list = [121, 2, 1, 1, 1, 1, 3, 3, 6, 3, 1, 2, 6, 2, 2, 15, 1, 1, 2, 2, 1]
elif conf['n_labeled_examples'] == 165:
split_list = [136, 2, 2, 1, 1, 1, 2, 4, 8, 3, 1, 2, 7, 2, 2, 18, 1, 1, 1, 3, 3]
### Cityscapes Dataset
elif config['dataset'] == 'cityscapes':
label_folder = 'datasets/cityscapes/segmentation/train'
if conf['n_labeled_examples'] == 372:
split_list = [42, 7, 26, 1, 2, 2, 1, 1, 19, 2, 5, 2, 1, 8, 1, 1, 1, 1, 1]
elif conf['n_labeled_examples'] == 186:
split_list = [45, 7, 28, 1, 2, 2, 1, 1, 20, 2, 5, 2, 1, 8, 1, 1, 1, 1, 1]
elif conf['n_labeled_examples'] == 93:
split_list = [38, 6, 22, 1, 2, 2, 1, 1, 17, 2, 5, 1, 1, 7, 1, 1, 1, 1, 1]
save_folder = os.path.join(model_folder, 'label_subcls_' + clustering_algorithm)
os.makedirs(save_folder, exist_ok=True)
feature_folder = os.path.join(model_folder, 'features')
subclasses = np.cumsum(np.asarray(split_list))
subclasses = np.insert(subclasses, 0, 0)
oldtime=datetime.datetime.now()
files = os.listdir(feature_folder)
list.sort(files)
feat_shape_list = []
label_shape_list = []
for i, file in enumerate(files):
feat = np.load(os.path.join(feature_folder, file))
feat_shape_list.append(feat.shape[-2:])
H, W = feat.shape[-2], feat.shape[-1]
if config['dataset'] == 'cityscapes':
label = np.asarray(Image.open(os.path.join(label_folder, file.replace('_leftImg8bit.npy', '_gtFine_labelTrainIds.png'))))
else:
label = np.asarray(Image.open(os.path.join(label_folder, file.replace('.npy', '.png'))))
label_shape_list.append(label.shape[-2:])
target = cv2.resize(label, (W,H), interpolation=cv2.INTER_NEAREST)
feat = feat.reshape(feat.shape[0],-1).transpose((1,0))
target = np.expand_dims(target.reshape(-1), axis=1)
if i==0:
feats = feat
targets = target
file_id = i * np.ones(target.shape)
else:
feats = np.vstack((feats, feat))
targets = np.vstack((targets, target))
file_id = np.vstack((file_id, i * np.ones(target.shape)))
newtime=datetime.datetime.now()
print('data_loading：%s'%(newtime-oldtime))
print(Counter(targets.reshape(-1).tolist()))
if clustering_algorithm == 'normal_kmeans':
targets_subcls = targets.copy()
for cls in np.unique(targets):
print('Parent class:', cls)
oldtime = datetime.datetime.now()
if cls < 255:
num_clusters = split_list[cls]
subcls = subclasses[cls]
if num_clusters == 1:
targets_subcls[targets==cls] = subcls
else:
subindex = np.where(targets==cls)[0]
subfeats = feats[subindex,:]
k_center = MiniBatchKMeans(n_clusters=num_clusters, random_state=0).fit(subfeats)
newtime = datetime.datetime.now()
print('KMeans：%s' % (newtime - oldtime))
lbls = k_center.labels_
for j in range(num_clusters):
targets_subcls[subindex[lbls==j]] = subcls + j
elif clustering_algorithm == 'balanced_kmeans':
targets_subcls = targets.copy()
for cls in np.unique(targets):
print('Parent class:', cls)
if cls < 255:
num_clusters = split_list[cls]
subcls = subclasses[cls]
if num_clusters == 1:
targets_subcls[targets == cls] = subcls
else:
subindex = np.where(targets == cls)[0]
subfeats = feats[subindex, :]
data_int16_x1000 = np.int16(subfeats * 1000)
np.savetxt(save_folder + '/subfeats_cls' + str(cls) + '_n' + str(num_clusters) + '.csv', data_int16_x1000,
fmt='%i', delimiter=',')
command = "regularized-k-means/build/regularized-k-means hard "+ save_folder + "/subfeats_cls" + str(cls) + "_n" + str(num_clusters) + ".csv " + str(num_clusters) + \
" -a " + save_folder + "/subfeats_cls" + str(cls) + "_n" + str(num_clusters) + "_hard_assignments -o"+ save_folder + "/subfeats_cls" + str(cls) + "_n" + str(num_clusters) + "_hard_summary.txt -t 20"
# print(command)
os.system(command)
lbls = np.loadtxt(save_folder + "/subfeats_cls" + str(cls) + "_n" + str(
num_clusters) + "_hard_assignments.csv", delimiter=',')
print(Counter(lbls.reshape(-1).tolist()))
for j in range(num_clusters):
targets_subcls[subindex[lbls == j]] = subcls + j
for i, file in enumerate(files):
tgt_subcls = targets_subcls[file_id==i]
feat_shape = feat_shape_list[i]
tgt_subcls = tgt_subcls.reshape(feat_shape)
H, W = label_shape_list[i]
tgt_subcls = cv2.resize(tgt_subcls, (W,H), interpolation=cv2.INTER_NEAREST)
Image.fromarray(tgt_subcls).save(os.path.join(save_folder, file.replace('.npy','.png')))
print(Counter(targets_subcls.reshape(-1).tolist()))
if __name__ == '__main__':
# PARSE THE ARGS
parser = argparse.ArgumentParser(description='PyTorch Training')
parser.add_argument('-c', '--config', default='configs/config.json', type=str,)
parser.add_argument('-ca', '--clustering_algorithm', default='balanced_kmeans', type=str,
help="Support 'balanced_kmeans' or 'normal_kmeans'")
args = parser.parse_args()
config = json.load(open(args.config))
main(config, args.clustering_algorithm)