-
Notifications
You must be signed in to change notification settings - Fork 4
/
val.py
142 lines (115 loc) · 4.02 KB
/
val.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
from __future__ import print_function
import argparse
import os
import shutil
import time
from tqdm import tqdm
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.optim
#import torch.utils.data
import torch.nn.functional as F
import torchvision.transforms as transforms
import torchvision.datasets as datasets
#from torchsummary import summary
import torchvision.models as models
# from models import *
from collections import OrderedDict
from torch.autograd import Variable
# import scipy as sp
from scipy import signal
from utils.utils import Normalize
from utils.utils import calc_scores
import logging
# import models.resnet as ResNet
import utils
import matplotlib.pyplot as plt
import numpy as np
# import cv2
import sys
from EvaluationMetrics.CCC_score import CCC_score
import math
from losses.CCC import CCC
import wandb
def validate(val_loader, audio_model, visual_model, model, criterion, epoch, cam):
# switch to evaluate mode
global Val_acc
global best_Val_acc
global best_Val_acc_epoch
model.eval()
audio_model.eval()
visual_model.eval()
cam.eval()
PrivateTest_loss = 0
correct = 0
total = 0
running_val_loss = 0
running_val_accuracy = 0
out = []
tar = []
#torch.cuda.synchronize()
#t7 = time.time()
for batch_idx, (visual_data, audiodata, labels) in tqdm(enumerate(val_loader),
total=len(val_loader), position=0, leave=True):
#if(batch_idx > 2):#int(65844/64)):
# break
#torch.cuda.synchronize()
#t8 = time.time()
#print('data loading time', t8-t7)
audiodata = audiodata.cuda()
visualdata = visual_data.cuda()
#torch.cuda.synchronize()
#t9 = time.time()
with torch.no_grad():
b, c, seq_t, subseq_t, h, w = visualdata.size()
#sub_seq_len = 16
#visualdata = visual_data.view(b, c, -1, sub_seq_len, h, w)
visual_feats = []
aud_feats = []
for i in range(visualdata.shape[0]):
vis_dat = visualdata[i, :, :, :,:,:].transpose(0,1)
visualfeat, _ = visual_model(vis_dat)
visual_feats.append(visualfeat)
aud_data = audiodata[i,:,:,:].unsqueeze(1)
audio_feat, audio_out = audio_model(aud_data)
aud_feats.append(audio_feat.squeeze(3))
visual_feat = torch.stack(visual_feats).squeeze(3).squeeze(3).squeeze(3)#.transpose(1,2)
audio_feat = torch.stack(aud_feats).squeeze(3)#.transpose(1,2)
#torch.cuda.synchronize()
#t8 = time.time()
#audio_feat, audio_out = audio_model(audiodata)
#audio_feat = audio_feat.squeeze(3)
#audio_feat, audio_out = audio_model(audiodata)
#visualfeat, visual_out = visual_model(visualdata)#.unsqueeze(0))
#visual_feat = visualfeat.squeeze(2).squeeze(2).squeeze(2)
#visual_feat = torch.max(visualfeat, dim = 2)[0].squeeze(2).squeeze(2)
#vis_data = visualdata.view(b*visualdata.shape[2], c, subseq_t ,h , w)
#visualfeatures, _ = visual_model(vis_data)
#visual_feat = visualfeatures.view(b, -1, visualfeatures.shape[1])
#aud_data = audiodata.view(audiodata.shape[0]*audiodata.shape[1], audiodata.shape[2], audiodata.shape[3]).unsqueeze(1)
#aud_feat, audio_out = audio_model(aud_data)
#audio_feat = aud_feat.view(b, -1, aud_feat.shape[1])
#print(audio_feat.shape)
#print(visual_feat.shape)
audio_feat_norm = F.normalize(audio_feat, p=2, dim=2, eps=1e-12)
visual_feat_norm = F.normalize(visual_feat, p=2, dim=2, eps=1e-12)
audio_attfeat, visual_attfeat = cam(audio_feat, visual_feat)
audiovisual_outs = model(audio_feat_norm, visual_feat_norm)
outputs = audiovisual_outs.view(-1, audiovisual_outs.shape[0]*audiovisual_outs.shape[1])
targets = labels.view(-1, labels.shape[0]).cuda()
val_loss = criterion(outputs, targets)
if batch_idx % 10 == 0:
wandb.log({"val_loss": val_loss})
out = np.concatenate([out, outputs.squeeze(0).detach().cpu().numpy()])
tar = np.concatenate([tar, targets.squeeze(0).detach().cpu().numpy()])
pred, tar = Normalize(out, tar)
if (len(tar) > 1):
Val_acc = CCC_score(tar, pred)
else:
Val_acc = 0
print("Val Accuracy")
wandb.log({"Val_acc": Val_acc})
print(Val_acc)
return val_loss, (Val_acc)