def converter(batch, device):
# The converted receives a batch of input samples any may modify it if
# necessary. In our case, we need to align the captions depending on if
# we are using LSTM layers of NStepLSTM layers in the model.
if args.rnn == 'lstm':
max_caption_length = args.max_caption_length
elif args.rnn == 'nsteplstm':
max_caption_length = None
else:
raise ValueError('Invalid RNN type.')
return datasets.converter(
batch, device, max_caption_length=max_caption_length)
def converter(batch, device):
# The converted receives a batch of input samples any may modify it if
# necessary. In our case, we need to align the captions depending on if
# we are using LSTM layers of NStepLSTM layers in the model.
if args.rnn == 'lstm':
max_caption_length = args.max_caption_length
elif args.rnn == 'nsteplstm':
max_caption_length = None
else:
raise ValueError('Invalid RNN type.')
return datasets.converter(
batch, device, max_caption_length=max_caption_length)