-
Notifications
You must be signed in to change notification settings - Fork 1
/
prediction.py
129 lines (105 loc) · 5.31 KB
/
prediction.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
import torch
import torch.nn as nn
from torch.autograd import Variable, Function
from torch.utils.data import Dataset, DataLoader
import torchvision
import torch.utils.data as data
import os
import os.path
import h5py
import numpy as np
import argparse
from thumos_dataset_precalc_feature import thumos_dataset_eval
from model import model
gpu_id=None
mode = 'test'
batch_size=1
frame_interval=5
frame_length=32
gen_feature_len = 12
workers=0
label_num = 21
feature_rgb_dir_test='data/rgb_feature_test_interval5.h5'
feature_flow_dir_test='data/flow_feature_test_interval5.h5'
label_dir_test = 'data/test_frame.txt'
label_videolevel_test = 'data/test_video_level_label.txt'
outfile_path='prediction.h5'
load_dir=None
load_dir='./checkpoints/ckp-genlen12-ep205-best.pt'
parser = argparse.ArgumentParser(description='Process some integers.')
parser.add_argument('--infer_steps', type=int, default=4,
help='number of multiple inference')
parser.add_argument('--feature_size', type=int, default=2048,
help='channel size of input feature')
args = parser.parse_args()
feature_size=args.feature_size
infer_steps=args.infer_steps
print(infer_steps)
model = model(feature_size, label_num)
model.cuda(gpu_id)
if(load_dir):
checkpoint = torch.load(load_dir)
model.load_state_dict(checkpoint['model'])
print("loaded checkpoint %s" % (load_dir))
if(mode=='test'):
outfile = h5py.File(outfile_path, 'w')
processed_video=[]
with open(label_videolevel_test, "r") as fp:
for string in fp:
string = string.split()
if(string[0] in processed_video):
continue
processed_video.append(string[0])
vidfile = string[0]
print(vidfile)
val_dataset = thumos_dataset_eval(feature_rgb_dir_test,feature_flow_dir_test, vidfile, label_dir_test, frame_length, frame_interval, gen_feature_len)
val_dataloader = DataLoader(
val_dataset,
batch_size=batch_size, shuffle=False,
num_workers=workers)
duration = val_dataset.label_seq.size(0)
print(val_dataset.label_seq.size())
dset_label = outfile.create_dataset(string[0]+'/label', (duration,label_num), maxshape=(duration,label_num), chunks=True, dtype=np.float32)
dset_label[:,:] = val_dataset.label_seq.numpy()
dset_pred = outfile.create_dataset(string[0]+'/pred', (duration,label_num), maxshape=(duration,label_num), chunks=True, dtype=np.float32)
dset_pred[:,0]=1.0
dset_weight = outfile.create_dataset(string[0]+'/weight', (duration,2), maxshape=(duration,2), chunks=True, dtype=np.float32)
dset_weight[:,1]=1.0
model.eval()
with torch.no_grad():
seg = []
seg_h = []
seg_decode_h = []
for i in range(0, infer_steps):
seg.append(-i * frame_length/infer_steps)
seg_h.append( torch.zeros(batch_size, 4096, requires_grad = True).cuda() )
seg_decode_h.append( torch.zeros(batch_size, 4096, requires_grad = True).cuda() )
for i, sample_batched in enumerate(val_dataloader):
X=sample_batched['data']
X=Variable(X)
X=X.cuda(gpu_id)
idx = sample_batched['idx']
prev_prob = torch.zeros(batch_size, label_num).cuda()
for seg_idx in range(0, frame_length):
current_feature = X[:,seg_idx,:]
infer_cnt = 0
for infer_idx in range(0,infer_steps):
if(seg[infer_idx] >= 0):
infer_cnt += 1
if(seg[infer_idx] == frame_length ):
seg[infer_idx] = 0
seg_h[infer_idx].zero_()
seg_decode_h[infer_idx].zero_()
y_pred, seg_decode_h[infer_idx], seg_h[infer_idx], weight = model.forward_unroll(current_feature, seg_decode_h[infer_idx], seg_h[infer_idx], prev_prob, gen_feature_len)
y_pred = nn.Softmax(dim=1)(y_pred)
y_pred = y_pred.cpu().numpy()
weight = weight.cpu().numpy()
for batch in range(0, y_pred.shape[0]):
if(idx[batch]+seg_idx <duration):
dset_pred[idx[batch]+seg_idx,:]+=y_pred[batch,:]
dset_weight[idx[batch]+seg_idx,:]=weight[batch,:]
seg[infer_idx] += 1
for batch in range(0, X.size(0)):
if(idx[batch]+seg_idx <duration):
dset_pred[idx[batch]+seg_idx,:]/=infer_cnt
prev_prob = torch.from_numpy(dset_pred[idx[0]+seg_idx,:]).view(1,-1).cuda()