TaskPaths object just provides paths to common directories that are used in different tasks, including NN training and inference. This paths are fixed because required directories are mounted as volume in Docker container.
class TaskPaths: def __init__(self, determine_in_project=True):
determine_in_projectis True, property
project_dirwill be set to path to (single) project in
Path to json file with input task settings.
Path to NN model (checkpoint), i.e. directory with stored model weights. It is a model:
to apply it — for NN inference task
to continue training since it or to copy weights from it — for NN training task
Content of the directory is entirely dependent on model implementation.
Path to directory where task results must be stored. It will be:
project in Supervisely format (possibly without images, with annotations only) — for NN inference task
directories with checkpoints — for NN training task
Path to input project in Supervisely format. It will be:
data to apply NN model on it — for NN inference task
training data — for NN training task
Path to directory where some additional (debug) data may be stored. It may be useful for running tasks locally while developing (implementing NNs or something else).