Training

Functions

janice_fine_tune

Fine tune an implementation using new data. This function can be used to adapt an algorithm to a new domain. It is optional and can return JANICE_NOT_IMPLEMENTED. Artifacts created from fine tuning should be stored on disk and will be loaded in a successive initialization of the API.

Signature

JANICE_EXPORT JaniceError janice_fine_tune(const JaniceMediaIterators* media,
                                           const JaniceDetectionsGroup* detections,
                                           int** labels,
                                           const char* output_prefix);

Thread Safety

This function is Thread Unsafe.

Parameters

Name Type Description
media const JaniceMediaIterators* A list of media objects to fine tune with. After the function call, each iterator in the array will exist in an undefined state. A user should call reset on each iterator before reusing them.
detections const JaniceDetectionsGroup* A collection of location information for objects in the fine tuning data. There must be the same number of sublists in this structure as there are elements in media. The tracks in the ith sublist of this structure give locations in the ith media object.
labels int** A list of lists of labels for objects in the fine tuning data. labels[i][j] should give the label for the jth track in the ith sublist of tracks.
output_prefix const char* A path to an existing directory with write access for the application. After successful fine tuning, this directory should be populated with all files necessary to initialize the API. Future calls to the API can use the fine tuned algorithm by passing output_prefix as the sdk_path parameter in janice_initialize.