-
Notifications
You must be signed in to change notification settings - Fork 3
/
generate_pseudo_label_arid_emu.py
101 lines (76 loc) · 3.48 KB
/
generate_pseudo_label_arid_emu.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
# from winreg import EnumValue
import torch
import pickle
import csv
import argparse
import pandas as pd
import numpy as np
parser = argparse.ArgumentParser()
parser.add_argument('--gen_label_list_path', type=str, default="./data/arid_unlabel.csv")
parser.add_argument('--gen_label_pkl_path_ssv2_best', type=str, default="./exp_adapt_bn/uniformer_b32_ssv2/aridx32x224x1x1.pkl")
parser.add_argument('--gen_label_pkl_path_uni_best', type=str, default="./exp_adapt_bn/uniformer_b32_k600/aridx32x224x1x1.pkl")
parser.add_argument('--gen_label_pkl_path_mvit_best', type=str, default="./exp_adapt_bn/mvit_b32_k600_dp0.3/aridx32x224x1x1.pkl")
parser.add_argument('--gen_label_pkl_path_slowfast_best', type=str, default="./exp_adapt_bn/sf32_k600/aridx32x224x1x1.pkl")
args = parser.parse_args()
def get_pseudo_label(cla,th,idx,ssv2_best_probs,uni_best_probs,mvit_best_probs,slowfast_best_probs):
ssv2_probs = ssv2_best_probs
ssv2_preds = ssv2_probs.argmax(1)
indx_7_6 = np.bitwise_or((ssv2_preds==6),(ssv2_preds==7))
other_indx = np.bitwise_xor(np.ones(len(ssv2_preds)).astype(np.bool),indx_7_6)
new_probs = np.zeros(ssv2_probs.shape)
uni_probs = uni_best_probs
new_probs[other_indx]=ssv2_probs[other_indx]*0.375+uni_probs[other_indx]*0.375+mvit_best_probs[other_indx]*0.1+slowfast_best_probs[other_indx]*0.1
new_probs[indx_7_6] = ssv2_probs[indx_7_6]*0.8 + uni_probs[indx_7_6]*0.2
probs = new_probs
new_preds = new_probs.argmax(1)
preds = []
idx_new = []
probs_new = []
for i,_ in enumerate(new_probs):
if new_probs[i].argmax() == cla and max(new_probs[i]) >= th:
preds.append(new_preds[i])
idx_new.append(idx[i])
probs_new.append(probs[i])
return preds,idx_new,probs_new
ssv2_best_probs = torch.Tensor(pickle.load(open(args.gen_label_pkl_path_ssv2_best, 'rb'))['video_preds']).softmax(-1).numpy()[3088:]
uni_best_probs = torch.Tensor(pickle.load(open(args.gen_label_pkl_path_uni_best, 'rb'))['video_preds']).softmax(-1).numpy()[3088:]
mvit_best_probs = torch.Tensor(pickle.load(open(args.gen_label_pkl_path_mvit_best, 'rb'))['video_preds']).softmax(-1).numpy()[3088:]
slowfast_best_probs = torch.Tensor(pickle.load(open(args.gen_label_pkl_path_slowfast_best, 'rb'))['video_preds']).softmax(-1).numpy()[3088:]
f=open(args.gen_label_list_path,"r")
idx = f.readlines()
f.close()
idx = [x.strip().split(",")[0] for x in idx]
real_csv_list = []
with open('./data/real_arid_unlabel.csv')as f:
f_csv = csv.reader(f)
for i, row in enumerate(f_csv):
real_csv_list.append(row)
real_label_cnt ={}
for i in range(0,11):
real_label_cnt[i] = 0
real = {}
for i,row in enumerate(real_csv_list):
real[row[0]]=row[1]
real_label_cnt[int(row[1])] += 1
cla = 10
check_l = {}
t_list = np.linspace(0, 1, 1001)
print("real_label_cnt",real_label_cnt[cla])
maxx = 0
for th in t_list:
pred,idx_x,prob = get_pseudo_label(cla,th,idx,ssv2_best_probs,uni_best_probs,mvit_best_probs,slowfast_best_probs)
prob = np.array(prob)
t_cnt = 0
for i,_ in enumerate(idx_x):
if int(real[idx_x[i]]) == int(pred[i]):
t_cnt+=1
ratio = t_cnt/len(pred)
f1 = (2*(len(pred)/real_label_cnt[cla])*(t_cnt/len(pred)))/(len(pred)/real_label_cnt[cla])+(t_cnt/len(pred))
if f1>maxx:
maxx=f1
print("th:",th)
print("s",len(pred))
print("total number of real_label = ",t_cnt)
print("acc = ",t_cnt/len(pred))
print("acc2 = ",len(pred)/real_label_cnt[cla])
print("F1",f1)