-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathevaluate_CFPRF.py
176 lines (171 loc) · 8.77 KB
/
evaluate_CFPRF.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
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
import torch,argparse
import numpy as np
from libs.dataloader.data_io import get_dataloader
from models.FDN import *
from models.PRN import *
from libs import tool
from libs.startup_config import set_random_seed
import glob
from tqdm import tqdm
import pandas as pd
import warnings
warnings.filterwarnings("ignore")
__author__ = "Junyan Wu"
__email__ = "[email protected]"
############EVAL CFPRF################
def Inference_CFPRF(eval_dlr, FDN_model, RPN_model, rso, device):
print("++++++++++++++++++inference++++++++++++++++++")
RPN_model.eval()
FDN_model.eval()
with torch.no_grad():
cp_dict={}
ver_dict={}
regver_dict={}
seg_score_dict={}
seg_tar_dict={}
for batch_x, filenames, batch_seg_label in tqdm(eval_dlr,ncols=50):
for fn in filenames:
cp_dict[fn]=np.array([])
ver_dict[fn]=np.array([])
regver_dict[fn]=np.array([])
if batch_x.shape[1]>340000:
continue
batch_x, batch_seg_label = batch_x.to(device), batch_seg_label.to(device)
batch_size = batch_x.size(0)
# inference FDN
seg_score,_, embs, last_hidden_states = FDN_model(batch_x)
seg_score, seg_target, seg_score_list,seg_target_list = prepare_segcon_target_ali(batch_seg_label, seg_score,rso)
seg_score_np=np.array([ss.data.cpu().numpy() for ss in seg_score_list])
# get coarse-gained proposal lists from FDN output scores with the initial confidence score set to 1
FDN_cp_list=[segscore2proposal(seg_score_np[idx],cp_fun=proposal_func,rso=rso)[1] for idx in range(batch_size)]
# save dict for FDN output
for idx,fn in enumerate(np.array(filenames)):
cp_dict[fn]=frame2second_proposal(FDN_cp_list[idx], rso=rso)
seg_score_dict[fn]=seg_score_np[idx][:,1].ravel()
seg_tar_dict[fn]=seg_target_list[idx].data.cpu().numpy().ravel()
if all(arr.size == 0 for arr in FDN_cp_list):
# If the coarse proposal list is null, continue
continue
# inference PRN
ver_pred, reg_pred, _, PRN_cp_list = RPN_model(last_hidden_states, FDN_cp_list, rso)
# VerH Decoder is applied to get the VerH proposal list
ver_label_pred= (ver_pred > 0.5).int().cpu().numpy()
cp_ver_list=decoder_ver(PRN_cp_list, ver_label_pred)
# save dict for VerH proposal
for idx,fn in enumerate(np.array(filenames)):
ver_dict[fn]=frame2second_proposal(cp_ver_list[idx],rso=rso)
# get the positive proposals
sampled_pos_inds = np.where(ver_label_pred == 1)[0]
PRN_cp_flatten = np.concatenate([arr for arr in PRN_cp_list if arr.size > 0]).reshape(-1, 3)
# Compute the proposal lengths
# Forgery segments smaller than two frames will not be used in the regression process
lengths = PRN_cp_flatten[sampled_pos_inds][:, 2] - PRN_cp_flatten[sampled_pos_inds][:, 1]
short_inds = np.where(lengths < 2)[0]
fp_pred_list,fp_short_list=[np.array([]),np.array([])],[np.array([]),np.array([])]
if short_inds.size>0: # only verification decoder applied
sampled_pos_short_inds=sampled_pos_inds[short_inds]
sampled_pos_inds = np.delete(sampled_pos_inds, short_inds) # remove the short inds
short_label_pred = np.zeros(len(ver_label_pred))
short_label_pred [sampled_pos_short_inds] = 1
fp_short_list=decoder_ver(PRN_cp_list, short_label_pred)
if sampled_pos_inds.size >0: # use regression and verification decoders
pos_label_pred = np.zeros(len(ver_label_pred))
pos_label_pred [sampled_pos_inds] = 1
fp_pred_flatten = decoder_reg(PRN_cp_flatten, reg_pred.data.cpu().numpy())
fp_pred_flatten[fp_pred_flatten[:, 2] == 0] = PRN_cp_flatten[fp_pred_flatten[:, 2] == 0]
fp_pred_list= decoder_ver2(fp_pred_flatten, pos_label_pred, FDN_cp_list)
# save dict for fine-grained (VerH+RegH) proposals
for fn, fp_short, fp_pred in zip(filenames, fp_short_list, fp_pred_list):
if fp_short.size == 0:
combined = fp_pred
elif fp_pred.size == 0:
combined = fp_short
else:
combined = vstack_twolist(fp_short, fp_pred)
regver_dict[fn] = frame2second_proposal(combined, rso=rso)
return seg_score_dict, seg_tar_dict, cp_dict, ver_dict, regver_dict
def FDN_performance_excel(seg_score_dict,seg_tar_dict):
Seer_value,Sacc,f1,pre,rec, Sauc=eval_PFD(seg_score_dict,seg_tar_dict)
savecontent=pd.DataFrame()
savecontent['Seer']=["%.2f"%float(Seer_value)]
savecontent['Sauc']=["%.2f"%float(Sauc)]
savecontent['P']=["%.2f"%float(pre)]
savecontent['R']=["%.2f"%float(rec)]
savecontent['F1']=["%.2f"%float(f1)]
savecontent['Sacc']=["%.2f"%float(Sacc)]
return savecontent
def PRN_performance_excel(gt_dict,proposal_dict):
mAP_value, ap_score, ar_score=eval_TFL(gt_dict, proposal_dict, detail=True)
savecontent=pd.DataFrame()
savecontent['mAP']=["%.2f"%float(mAP_value*100)]
for ap in ap_score:
savecontent['AP@'+str(ap)]=["%.2f"%float(ap_score[ap]*100)]
for ar in sorted(ar_score):
savecontent['AR@'+str(ar)]=["%.2f"%float(ar_score[ar]*100)]
return savecontent
if __name__ == '__main__':
parser = argparse.ArgumentParser('python evaluate_CFPRF.py --dn HAD/PS/LAVDF')
parser.add_argument('--seed', type=int, default=1234)
parser.add_argument('--save_path', type=str, default="./result/") # ['HAD','PS','LAVDF']
parser.add_argument('--dn', type=str, default="HAD") # ['HAD','PS','LAVDF']
parser.add_argument('--seql', type=int, default=1070)
parser.add_argument('--rso', type=int, default=20)
parser.add_argument('--glayer', type=int, default=1)
parser.add_argument('--eval', action='store_true', default=True)
args = parser.parse_args()
set_random_seed(args.seed)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
"""loading dataset"""
test_gt_dict, test_dlr=get_dataloader(batch_size=1,part="test",dn=args.dn,rso=args.rso)
"""loading FDN model"""
FDN_model = CFPRF_FDN(seq_len=args.seql, gmlp_layers=args.glayer).to(device)
"""
./checkpoints
├── 1FDN_HAD.pth
├── 1FDN_LAVDF.pth
├── 1FDN_PS.pth
├── 2PRN_HAD.pth
├── 2PRN_LAVDF.pth
├── 2PRN_PS.pth
"""
FDN_checkpoint="./checkpoints/1FDN_%s.pth"%(args.dn)
FDN_model.load_state_dict(torch.load(FDN_checkpoint))
"""loading PRN model"""
RPN_model = CFPRF_PRN(device=device).to(device)
PRN_modelpath="./checkpoints/2PRN_%s.pth"%(args.dn)
RPN_model.load_state_dict(torch.load(PRN_modelpath))
"""makedir"""
dict_save_path=os.path.join(args.save_path,'dict/%s_'%(args.dn))
csv_save_path=os.path.join(args.save_path,'pd/%s_'%(args.dn))
os.makedirs(os.path.dirname(dict_save_path),exist_ok=True)
os.makedirs(os.path.dirname(csv_save_path),exist_ok=True)
###########INFERENCE#############
if args.eval:
seg_score_dict, seg_tar_dict, cp_dict, ver_dict, regver_dict=Inference_CFPRF(test_dlr, FDN_model, RPN_model, args.rso, device)
writenpy(dict_save_path+'seg_tar_dict.npy',seg_tar_dict)
writenpy(dict_save_path+'seg_score_dict.npy',seg_score_dict)
writenpy(dict_save_path+'cp_dict.npy',cp_dict)
writenpy(dict_save_path+'ver_dict.npy',ver_dict)
writenpy(dict_save_path+'regver_dict.npy',regver_dict)
else:
seg_score_dict=readnpy(dict_save_path+'seg_score_dict.npy',seg_score_dict)
seg_tar_dict=readnpy(dict_save_path+'seg_tar_dict.npy',seg_tar_dict)
cp_dict=readnpy(dict_save_path+'cp_dict.npy',cp_dict)
ver_dict=readnpy(dict_save_path+'ver_dict.npy',ver_dict)
regver_dict=readnpy(dict_save_path+'regver_dict.npy',regver_dict)
###########PFD#############
savecontent0=FDN_performance_excel(seg_score_dict,seg_tar_dict)
savecontent0.to_csv(csv_save_path+'PFD_results.csv')
print("PFD",savecontent0)
###########TFL_CP#############
savecontent1=PRN_performance_excel(test_gt_dict,cp_dict)
savecontent1.to_csv(csv_save_path+'TFL_CP_results.csv')
print("CP",savecontent1)
###########TFL_VER#############
savecontent2=PRN_performance_excel(test_gt_dict,ver_dict)
savecontent2.to_csv(csv_save_path+'TFL_VER_results.csv')
print("VER",savecontent2)
###########TFL_RP#############
savecontent3=PRN_performance_excel(test_gt_dict,regver_dict)
savecontent3.to_csv(csv_save_path+'TFL_RP_results.csv')
print("RP",savecontent3)