-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest.py
146 lines (130 loc) · 7.57 KB
/
test.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
import json
import os
import argparse
import torch
from torch import nn
from torch.autograd import Variable
from torch.utils.data import DataLoader
from models import EncoderRNN, DecoderRNN, S2VTAttModel, S2VTModel
from dataloader import VideoDataset
import misc.utils as utils
from misc.cocoeval import suppress_stdout_stderr, COCOScorer
import numpy as np
from pandas.io.json import json_normalize
opt1 = json.load(open('./save_model/vatex_msrvtt_res152_i3d/opt_info.json', 'r'))
opt2 = json.load(open('./save_model/vatex_msrvtt_res152/opt_info.json', 'r'))
opt3 = json.load(open('./save_model/vatex_msrvtt_msvd_res152_i3d/opt_info.json', 'r'))
dataset1 = VideoDataset(opt1, "test")
dataset2 = VideoDataset(opt2, "test")
dataset3 = VideoDataset(opt3, "test")
opt1["vocab_size"] = 27170
opt1["seq_length"] = 28
opt2["vocab_size"] = 27170
opt2["seq_length"] = 28
opt3["vocab_size"] = 28620
opt3["seq_length"] = 28
encoder = EncoderRNN(opt1["dim_vid"], opt1["dim_hidden"], bidirectional=bool(opt1["bidirectional"]),
input_dropout_p=opt1["input_dropout_p"], rnn_dropout_p=opt1["rnn_dropout_p"])
decoder = DecoderRNN(opt1["vocab_size"], opt1["max_len"], opt1["dim_hidden"], opt1["dim_word"],
input_dropout_p=opt1["input_dropout_p"],
rnn_dropout_p=opt1["rnn_dropout_p"], bidirectional=bool(opt1["bidirectional"]))
model1 = S2VTAttModel(encoder, decoder).cuda()
model1.load_state_dict({k.replace('module.',''):v for k,v in torch.load('./save_model/vatex_msrvtt_res152_i3d/model_320.pth').items()})
encoder = EncoderRNN(opt1["dim_vid"], opt1["dim_hidden"], bidirectional=bool(opt1["bidirectional"]),
input_dropout_p=opt1["input_dropout_p"], rnn_dropout_p=opt1["rnn_dropout_p"])
decoder = DecoderRNN(opt1["vocab_size"], opt1["max_len"], opt1["dim_hidden"], opt1["dim_word"],
input_dropout_p=opt1["input_dropout_p"],
rnn_dropout_p=opt1["rnn_dropout_p"], bidirectional=bool(opt1["bidirectional"]))
model2 = S2VTAttModel(encoder, decoder).cuda()
model2.load_state_dict({k.replace('module.',''):v for k,v in torch.load('./save_model/vatex_msrvtt_res152_i3d/model_300.pth').items()})
encoder = EncoderRNN(opt2["dim_vid"], opt2["dim_hidden"], bidirectional=bool(opt2["bidirectional"]),
input_dropout_p=opt2["input_dropout_p"], rnn_dropout_p=opt2["rnn_dropout_p"])
decoder = DecoderRNN(opt2["vocab_size"], opt2["max_len"], opt2["dim_hidden"], opt2["dim_word"],
input_dropout_p=opt2["input_dropout_p"],
rnn_dropout_p=opt2["rnn_dropout_p"], bidirectional=bool(opt2["bidirectional"]))
model3 = S2VTAttModel(encoder, decoder).cuda()
model3.load_state_dict({k.replace('module.',''):v for k,v in torch.load('./save_model/vatex_msrvtt_res152/model_320.pth').items()})
encoder = EncoderRNN(opt2["dim_vid"], opt2["dim_hidden"], bidirectional=bool(opt2["bidirectional"]),
input_dropout_p=opt2["input_dropout_p"], rnn_dropout_p=opt2["rnn_dropout_p"])
decoder = DecoderRNN(opt2["vocab_size"], opt2["max_len"], opt2["dim_hidden"], opt2["dim_word"],
input_dropout_p=opt2["input_dropout_p"],
rnn_dropout_p=opt2["rnn_dropout_p"], bidirectional=bool(opt2["bidirectional"]))
model4 = S2VTAttModel(encoder, decoder).cuda()
model4.load_state_dict({k.replace('module.',''):v for k,v in torch.load('./save_model/vatex_msrvtt_res152/model_280.pth').items()})
encoder = EncoderRNN(opt3["dim_vid"], opt3["dim_hidden"], bidirectional=bool(opt3["bidirectional"]),
input_dropout_p=opt3["input_dropout_p"], rnn_dropout_p=opt3["rnn_dropout_p"])
decoder = DecoderRNN(opt3["vocab_size"], opt3["max_len"], opt3["dim_hidden"], opt3["dim_word"],
input_dropout_p=opt3["input_dropout_p"],
rnn_dropout_p=opt3["rnn_dropout_p"], bidirectional=bool(opt3["bidirectional"]))
model5 = S2VTAttModel(encoder, decoder).cuda()
model5.load_state_dict({k.replace('module.',''):v for k,v in torch.load('./save_model/vatex_msrvtt_msvd_res152_i3d/model_320.pth').items()})
encoder = EncoderRNN(opt3["dim_vid"], opt3["dim_hidden"], bidirectional=bool(opt3["bidirectional"]),
input_dropout_p=opt3["input_dropout_p"], rnn_dropout_p=opt3["rnn_dropout_p"])
decoder = DecoderRNN(opt3["vocab_size"], opt3["max_len"], opt3["dim_hidden"], opt3["dim_word"],
input_dropout_p=opt3["input_dropout_p"],
rnn_dropout_p=opt3["rnn_dropout_p"], bidirectional=bool(opt3["bidirectional"]))
model6 = S2VTAttModel(encoder, decoder).cuda()
model6.load_state_dict({k.replace('module.',''):v for k,v in torch.load('./save_model/vatex_msrvtt_msvd_res152_i3d/model_200.pth').items()})
model1.eval()
model2.eval()
model3.eval()
model4.eval()
model5.eval()
model6.eval()
results = {}
results['result'] = []
results['version'] = "VERSION 1.2"
results['external_data'] = {}
results['external_data']['used'] = 'true'
results['external_data']['details'] = 'We used VATEX and MSRVTT and MSVD datasets to train the model'
test_list = os.listdir('./data/test/')
for i in test_list:
v = {}
fc_feat = []
fc_feat.append(np.load('./data/test_res152_20_feats/' + i[:-4] +'.npy'))
fc_feat = np.concatenate(fc_feat, axis=1)
c3d_feat = np.load('./data/test_i3d_feats/' + i[:-4] +'.npy')
c3d_feat = c3d_feat.reshape(-1, 1024)
fc_feat = np.concatenate((fc_feat, c3d_feat), axis=1)
feat = torch.from_numpy(fc_feat).type(torch.FloatTensor)
feat = feat.cuda()
feat = feat.reshape((1,-1,3072))
seq_probs1, seq_preds1 = model1(feat, mode='inference', opt=opt1)
sents1 = utils.decode_sequence(dataset1.get_vocab(), seq_preds1)
seq_probs2, seq_preds2 = model2(feat, mode='inference', opt=opt1)
sents2 = utils.decode_sequence(dataset1.get_vocab(), seq_preds2)
if seq_probs1.sum() > seq_probs2.sum():
sent = sents1
else:
sent = sents2
v['video_id'] = i[:i.find('.')]
v['caption'] = sent[0] + '.'
results['result'].append(v)
for num,i in enumerate(test_list):
fc_feat = []
fc_feat.append(np.load('./data/test_res152_40_feats/' + i[:-4] +'.npy'))
fc_feat = np.concatenate(fc_feat, axis=1)
feat = torch.from_numpy(fc_feat).type(torch.FloatTensor)
feat = feat.cuda()
feat = feat.reshape((1,-1,2048))
seq_probs3, seq_preds3 = model3(feat, mode='inference', opt=opt2)
sents3 = utils.decode_sequence(dataset2.get_vocab(), seq_preds3)
seq_probs4, seq_preds4 = model4(feat, mode='inference', opt=opt2)
sents4 = utils.decode_sequence(dataset2.get_vocab(), seq_preds4)
results['result'][num]['caption'] += sents3[0] + '.' + sents4[0] + '.'
for num,i in enumerate(test_list):
fc_feat = []
fc_feat.append(np.load('./data/test_res152_20_feats/' + i[:-4] +'.npy'))
fc_feat = np.concatenate(fc_feat, axis=1)
c3d_feat = np.load('./data/test_i3d_feats/' + i[:-4] +'.npy')
c3d_feat = c3d_feat.reshape(-1, 1024)
fc_feat = np.concatenate((fc_feat, c3d_feat), axis=1)
feat = torch.from_numpy(fc_feat).type(torch.FloatTensor)
feat = feat.cuda()
feat = feat.reshape((1,-1,3072))
seq_probs5, seq_preds5 = model5(feat, mode='inference', opt=opt3)
sents5 = utils.decode_sequence(dataset3.get_vocab(), seq_preds5)
seq_probs6, seq_preds6 = model6(feat, mode='inference', opt=opt3)
sents6 = utils.decode_sequence(dataset3.get_vocab(), seq_preds6)
results['result'][num]['caption'] += sents5[0] + '.' + sents6[0] + '.'
json.dump(results, open('submit/result.json', 'w'))