def __getitem__(self, index):
cap_id = self.cap_ids[index]
video_id = getVideoId(cap_id)
# video
frame_list = self.video2frames[video_id]
frame_vecs = []
for frame_id in frame_list:
frame_vecs.append(self.visual_feat.read_one(frame_id))
frames_tensor = torch.Tensor(frame_vecs)
# text
caption = self.captions[cap_id]
if self.bow2vec is not None:
cap_bow = self.bow2vec.mapping(caption)
if cap_bow is None:
cap_bow = torch.zeros(self.bow2vec.ndims)
else:
cap_bow = torch.Tensor(cap_bow)
else:
cap_bow = None
if self.vocab is not None:
tokens = clean_str(caption)
caption = []
caption.append(self.vocab('<start>'))
caption.extend([self.vocab(token) for token in tokens])
caption.append(self.vocab('<end>'))
cap_tensor = torch.Tensor(caption)
else:
cap_tensor = None
return frames_tensor, cap_tensor, cap_bow, index, cap_id, video_id
def __getitem__(self, index):
cap_id = self.cap_ids[index]
caption = self.captions[cap_id]
if self.bow2vec is not None:
cap_bow = self.bow2vec.mapping(caption)
if cap_bow is None:
cap_bow = torch.zeros(self.bow2vec.ndims)
else:
cap_bow = torch.Tensor(cap_bow)
else:
cap_bow = None
if self.vocab is not None:
tokens = clean_str(caption)
caption = []
caption.append(self.vocab('<start>'))
caption.extend([self.vocab(token) for token in tokens])
caption.append(self.vocab('<end>'))
cap_tensor = torch.Tensor(caption)
else:
cap_tensor = None
return cap_tensor, cap_bow, index, cap_id
def preprocess(self, query, clear):
if clear:
words = clean_str(query)
else:
words = query.strip().split()
return words