def forward(self, sample: Any):
sample = self.pre_tensor_transform(sample)
sample = self.to_tensor_transform(sample)
if self.assert_contains_tensor:
if not _contains_any_tensor(sample):
raise MisconfigurationException(
"When ``to_tensor_transform`` is overriden, "
"``DataPipeline`` expects the outputs to be ``tensors``")
sample = self.post_tensor_transform(sample)
return sample
def pre_tensor_transform(self, samples: Any) -> Any:
if _contains_any_tensor(samples):
return samples
if isinstance(samples, str):
samples = [samples]
if isinstance(samples, (list, tuple)) and all(
isinstance(p, str) for p in samples):
outputs = []
for sample in samples:
outputs.append(pil_loader(sample))
return outputs
raise MisconfigurationException(
"The samples should either be a tensor, a list of paths or a path."
)
def forward(self, sample: Any) -> Any:
with self._current_stage_context:
with self._pre_tensor_transform_context:
sample = self.pre_tensor_transform(sample)
self.callback.on_pre_tensor_transform(sample, self.stage)
with self._to_tensor_transform_context:
sample = self.to_tensor_transform(sample)
self.callback.on_to_tensor_transform(sample, self.stage)
if self.assert_contains_tensor:
if not _contains_any_tensor(sample):
raise MisconfigurationException(
"When ``to_tensor_transform`` is overriden, "
"``DataPipeline`` expects the outputs to be ``tensors``"
)
with self._post_tensor_transform_context:
sample = self.post_tensor_transform(sample)
self.callback.on_post_tensor_transform(sample, self.stage)
return sample