-
Notifications
You must be signed in to change notification settings - Fork 37
/
msrvtt_dataloader.py
135 lines (115 loc) · 4.08 KB
/
msrvtt_dataloader.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
from __future__ import absolute_import
from __future__ import division
from __future__ import unicode_literals
from __future__ import print_function
import torch as th
from torch.utils.data import Dataset
import pickle
import torch.nn.functional as F
import numpy as np
import re
import pandas as pd
from collections import defaultdict
from torch.utils.data.dataloader import default_collate
import json
import random
class MSRVTT_DataLoader(Dataset):
"""MSRVTT dataset loader."""
def __init__(
self,
csv_path,
features_path,
we,
we_dim=300,
max_words=30,
):
"""
Args:
"""
self.data = pd.read_csv(csv_path)
self.features = th.load(features_path)
self.we = we
self.we_dim = we_dim
self.max_words = max_words
self.max_video = 30
def __len__(self):
return len(self.data)
def custom_collate(self, batch):
return default_collate(batch)
def _zero_pad_tensor(self, tensor, size):
if len(tensor) >= size:
return tensor[:size]
else:
zero = np.zeros((size - len(tensor), self.we_dim), dtype=np.float32)
return np.concatenate((tensor, zero), axis=0)
def _tokenize_text(self, sentence):
w = re.findall(r"[\w']+", str(sentence))
return w
def _words_to_we(self, words):
words = [word for word in words if word in self.we.vocab]
if words:
we = self._zero_pad_tensor(self.we[words], self.max_words)
return th.from_numpy(we)
else:
return th.zeros(self.max_words, self.we_dim)
def __getitem__(self, idx):
vid = self.data['video_id'].values[idx]
sentence = self.data['sentence'].values[idx]
caption = self._words_to_we(self._tokenize_text(sentence))
feat_2d = F.normalize(self.features['2d'][vid].float(), dim=0)
feat_3d = F.normalize(self.features['3d'][vid].float(), dim=0)
video = th.cat((feat_2d, feat_3d))
return {'video': video, 'text': caption, 'video_id': vid}
return {'2d': feat_2d, '3d': feat_3d, 'text': caption, 'video_id': vid}
class MSRVTT_TrainDataLoader(Dataset):
"""MSRVTT train dataset loader."""
def __init__(
self,
csv_path,
json_path,
features_path,
we,
we_dim=300,
max_words=30,
):
"""
Args:
"""
self.csv = pd.read_csv(csv_path)
self.data = json.load(open(json_path, 'r'))
self.features = th.load(features_path)
self.we = we
self.we_dim = we_dim
self.max_words = max_words
self.sentences = defaultdict(list)
for s in self.data['sentences']:
self.sentences[s['video_id']].append(s['caption'])
def __len__(self):
return len(self.csv)
def custom_collate(self, batch):
return default_collate(batch)
def _zero_pad_tensor(self, tensor, size):
if len(tensor) >= size:
return tensor[:size]
else:
zero = np.zeros((size - len(tensor), self.we_dim), dtype=np.float32)
return np.concatenate((tensor, zero), axis=0)
def _tokenize_text(self, sentence):
w = re.findall(r"[\w']+", str(sentence))
return w
def _words_to_we(self, words):
words = [word for word in words if word in self.we.vocab]
if words:
we = self._zero_pad_tensor(self.we[words], self.max_words)
return th.from_numpy(we)
else:
return th.zeros(self.max_words, self.we_dim)
def __getitem__(self, idx):
vid = self.csv['video_id'].values[idx]
rind = random.randint(0, len(self.sentences[vid]) - 1)
sentence = self.sentences[vid][rind]
feat_2d = F.normalize(self.features['2d'][vid].float(), dim=0)
feat_3d = F.normalize(self.features['3d'][vid].float(), dim=0)
video = th.cat((feat_2d, feat_3d))
caption = self._words_to_we(self._tokenize_text(sentence))
return {'video': video, 'text': caption}