Specifying different video file path on remote machine

[tomhagley6]

Hi, I'm trying to label on one machine and train my model on a different one. This means I need to specify a different video path during training, but I don't seem to be able to. The template job profiles don't have an option for this, and I've tried:

trainer = sleap.nn.training.TopDownMultiClassModelTrainer.from_config(
cfg, training_labels=labels, video_search_paths=videos

But still get the same FileNotFoundError: Could not find filename

Any help would be appreciated. Thanks!

[roomrys]

Hi @tomhagley6,

Well, I think we have found ourselves a bit of a logical error in SLEAP. I traced where we use video_search_paths when creating a Trainer via from_config and ended up in our DataReaders.from_labels method where I discovered that the video_search_paths are only used if the training_labels argument is a path to an slp (not the already loaded Labels object):

sleap/sleap/nn/training.py

Lines 155 to 171 in 833c2d5

	@classmethod
	def from_labels(
	cls,
	training: Union[Text, Labels],
	validation: Union[Text, Labels, float],
	test: Optional[Union[Text, Labels]] = None,
	video_search_paths: Optional[List[Text]] = None,
	labels_config: Optional[LabelsConfig] = None,
	update_config: bool = False,
	with_track_only: bool = False,
	) -> "DataReaders":
	"""Create data readers from `Labels` datasets as data providers."""
	if isinstance(training, str):
	logger.info(f"Loading training labels from: {training}")
	training: Labels = sleap.load_file(
	training, search_paths=video_search_paths
	)

and while we update the field in the training config for search_path_hints (if the Labels object is already loaded),

sleap/sleap/nn/training.py

Lines 132 to 142 in 833c2d5

	# Update the config fields with arguments (if not a full Labels instance).
	if update_config:
	if isinstance(training, Text):
	labels_config.training_labels = training
	if isinstance(validation, Text):
	labels_config.validation_labels = validation
	elif isinstance(validation, float):
	labels_config.validation_fraction = validation
	if isinstance(test, Text):
	labels_config.test_labels = test
	labels_config.search_path_hints = video_search_paths

I cannot seem to find where we actually use this pretty useful search_path_hints field. I am converting this discussion to an issue to get this resolved.

In the meantime, could you try passing in the path to the slp to training_labels instead of the Labels object and let us know how that works.

Thanks,
Liezl