def forward(self, x, fileinfo):
filenames = [nii_seq_tk.parse_filename(y) for y in fileinfo]
datalength = [nii_seq_tk.parse_length(y) for y in fileinfo]
if self.training:
feature_vec = self._compute_embedding(x, datalength)
scores = self._compute_score(feature_vec)
# target
target = self._get_target(filenames)
target_vec = torch.tensor(target,
device=x.device,
dtype=torch.long)
target_vec = target_vec.repeat(self.v_submodels)
return [scores, target_vec]
else:
feature_vec = self._compute_embedding(x, datalength)
scores = self._compute_score(feature_vec, True)
target = self._get_target(filenames)
print("Output, %s, %d, %f" %
(filenames[0], target[0], scores.mean()))
# don't write output score as a single file
return None
def forward(self, x, fileinfo):
#with torch.no_grad():
# vad_waveform = self.m_vad(x.squeeze(-1))
# vad_waveform = self.m_vad(torch.flip(vad_waveform, dims=[1]))
# if vad_waveform.shape[-1] > 0:
# x = torch.flip(vad_waveform, dims=[1]).unsqueeze(-1)
# else:
# pass
filenames = [nii_seq_tk.parse_filename(y) for y in fileinfo]
datalength = [nii_seq_tk.parse_length(y) for y in fileinfo]
if self.training:
feature_vec = self._compute_embedding(x, datalength)
scores = self._compute_score(feature_vec)
# target
target = self._get_target(filenames)
target_vec = torch.tensor(target, device=x.device).long()
target_vec = target_vec.repeat(self.v_submodels)
return [scores, target_vec, True]
else:
feature_vec = self._compute_embedding(x, datalength)
score = self._compute_score(feature_vec, True)
target = self._get_target(filenames)
print("Output, %s, %d, %f" %
(filenames[0], target[0], score.mean()))
# don't write output score as a single file
return None
