-
Notifications
You must be signed in to change notification settings - Fork 1
/
ensemble.py
120 lines (108 loc) · 5.3 KB
/
ensemble.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
import argparse
import pickle
import os
import numpy as np
from tqdm import tqdm
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--dataset',
required=True,
choices={'ntu/xsub', 'ntu/xview', 'ntu120/xsub', 'ntu120/xset', 'NW-UCLA'},
help='the work folder for storing results') # 选择数据集
parser.add_argument('--alpha',
default=1,
help='weighted summation',
type=float) # 设置默认的加权权重
parser.add_argument('--joint-dir',
help='Directory containing "epoch1_test_score.pkl" for joint eval results') # 关节点测试结果
parser.add_argument('--bone-dir',
help='Directory containing "epoch1_test_score.pkl" for bone eval results') # 骨架测试结果
parser.add_argument('--joint-motion-dir', default=None) # 关节运动测试结果
parser.add_argument('--bone-motion-dir', default=None) # 骨骼运动测试结果
arg = parser.parse_args() # 解析参数
dataset = arg.dataset # 获取数据集类型
if 'UCLA' in arg.dataset:
label = []
with open('./data/' + 'NW-UCLA/' + '/val_label.pkl', 'rb') as f:
data_info = pickle.load(f)
for index in range(len(data_info)):
info = data_info[index]
label.append(int(info['label']) - 1)
elif 'ntu120' in arg.dataset:
if 'xsub' in arg.dataset:
npz_data = np.load('./data/' + 'ntu120/' + 'NTU120_CSub.npz')
label = np.where(npz_data['y_test'] > 0)[1]
elif 'xset' in arg.dataset:
npz_data = np.load('./data/' + 'ntu120/' + 'NTU120_CSet.npz')
label = np.where(npz_data['y_test'] > 0)[1]
elif 'ntu' in arg.dataset:
if 'xsub' in arg.dataset:
npz_data = np.load('./data/' + 'ntu/' + 'NTU60_CS.npz')
# label中包含了L个动作所属的动作类别
label = np.where(npz_data['y_test'] > 0)[1] # shape: (L,60),60代表的是有60个类别的动作，L代表的是每个动作，(L,60)中为1的部分代表的是该动作属于的动作类别
elif 'xview' in arg.dataset:
npz_data = np.load('./data/' + 'ntu/' + 'NTU60_CV.npz')
label = np.where(npz_data['y_test'] > 0)[1]
else:
raise NotImplementedError
with open(os.path.join(arg.joint_dir, 'epoch1_test_score.pkl'), 'rb') as r1:
r1 = list(pickle.load(r1).items())
with open(os.path.join(arg.bone_dir, 'epoch1_test_score.pkl'), 'rb') as r2:
r2 = list(pickle.load(r2).items())
if arg.joint_motion_dir is not None:
with open(os.path.join(arg.joint_motion_dir, 'epoch1_test_score.pkl'), 'rb') as r3:
r3 = list(pickle.load(r3).items())
if arg.bone_motion_dir is not None:
with open(os.path.join(arg.bone_motion_dir, 'epoch1_test_score.pkl'), 'rb') as r4:
r4 = list(pickle.load(r4).items())
right_num = total_num = right_num_5 = 0
if arg.joint_motion_dir is not None and arg.bone_motion_dir is not None:
arg.alpha = [0.6, 0.6, 0.4, 0.4]
csv_f = np.zeros(60,60)
# len(label)代表的是动作的数量
for i in tqdm(range(len(label))):
l = label[i] # label[i]表示第i个动作属于的动作类别
_, r11 = r1[i] # _代表的是r1中测试的第几个动作； r11表示的是该动作属于某个动作的分数大小
_, r22 = r2[i]
_, r33 = r3[i]
_, r44 = r4[i]
# 将每个流中某个动作属于某个类别的测试分数分别乘以该流所对应的系数，再相加
r = r11 * arg.alpha[0] + r22 * arg.alpha[1] + r33 * arg.alpha[2] + r44 * arg.alpha[3]
# argsort()返回的是列表中元素从小到大排列时，每个元素在原列表中的位置索引
rank_5 = r.argsort()[-5:]
right_num_5 += int(int(l) in rank_5)
r = np.argmax(r) # 返回最大值在原列表中对应位置的索引
right_num += int(r == int(l))
total_num += 1
acc = right_num / total_num
acc5 = right_num_5 / total_num
elif arg.joint_motion_dir is not None and arg.bone_motion_dir is None:
arg.alpha = [0.6, 0.6, 0.4]
for i in tqdm(range(len(label))):
l = label[:, i]
_, r11 = r1[i]
_, r22 = r2[i]
_, r33 = r3[i]
r = r11 * arg.alpha[0] + r22 * arg.alpha[1] + r33 * arg.alpha[2]
rank_5 = r.argsort()[-5:]
right_num_5 += int(int(l) in rank_5)
r = np.argmax(r)
right_num += int(r == int(l))
total_num += 1
acc = right_num / total_num
acc5 = right_num_5 / total_num
else:
for i in tqdm(range(len(label))):
l = label[i]
_, r11 = r1[i]
_, r22 = r2[i]
r = r11 + r22 * arg.alpha
rank_5 = r.argsort()[-5:]
right_num_5 += int(int(l) in rank_5)
r = np.argmax(r)
right_num += int(r == int(l))
total_num += 1
acc = right_num / total_num
acc5 = right_num_5 / total_num
print('Top1 Acc: {:.4f}%'.format(acc * 100))
print('Top5 Acc: {:.4f}%'.format(acc5 * 100))